mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-24 03:00:14 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

imgproc: add IntelligentScissors

Browse Source
This commit is contained in:
Alexander Alekhin 2020-12-16 00:53:52 +00:00
parent 84676fefe3
commit 3eea3dd46b
14 changed files with 1637 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

127
doc/js_tutorials/js_assets/js_intelligent_scissors.html Normal file
View File

@ -0,0 +1,127 @@
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<title>Intelligent Scissors Example</title>
<link href="js_example_style.css" rel="stylesheet" type="text/css" />
</head>
<body>
<h2>Intelligent Scissors Example</h2>
<p>
Click <b>Start</b> button to launch the code below.<br>
Then click on image to pick source point. After that you can hover mouse pointer over canvas to specify target point candidate.<br>
You can change the code in the &lt;textarea&gt; to investigate more. You can choose another image (need to "Stop" first).
</p>
<div>
<div class="control"><button id="tryIt" disabled>Start</button> <button id="stopIt" disabled>Stop</button></div>
<textarea class="code" rows="20" cols="100" id="codeEditor" spellcheck="false">
</textarea>
<p class="err" id="errorMessage"></p>
</div>
<div id="inputParams">
<div class="caption">canvasInput <input type="file" id="fileInput" name="file" accept="image/*" /></div>
<canvas id="canvasInput"></canvas>
</div>
<div id="result" style="display:none">
<canvas id="canvasOutput"></canvas>
</div>
<script src="utils.js" type="text/javascript"></script>
<script id="codeSnippet" type="text/code-snippet">
let src = cv.imread('canvasInput');
//cv.resize(src, src, new cv.Size(1024, 1024));
cv.imshow('canvasOutput', src);
let tool = new cv.segmentation_IntelligentScissorsMB();
tool.setEdgeFeatureCannyParameters(32, 100);
tool.setGradientMagnitudeMaxLimit(200);
tool.applyImage(src);
let hasMap = false;
let canvas = document.getElementById('canvasOutput');
canvas.addEventListener('click', e => {
let startX = e.offsetX, startY = e.offsetY; console.log(startX, startY);
if (startX < src.cols && startY < src.rows)
{
console.time('buildMap');
tool.buildMap(new cv.Point(startX, startY));
console.timeEnd('buildMap');
hasMap = true;
}
});
canvas.addEventListener('mousemove', e => {
let x = e.offsetX, y = e.offsetY; //console.log(x, y);
let dst = src.clone();
if (hasMap && x >= 0 && x < src.cols && y >= 0 && y < src.rows)
{
let contour = new cv.Mat();
tool.getContour(new cv.Point(x, y), contour);
let contours = new cv.MatVector();
contours.push_back(contour);
let color = new cv.Scalar(0, 255, 0, 255); // RGBA
cv.polylines(dst, contours, false, color, 1, cv.LINE_8);
contours.delete(); contour.delete();
}
cv.imshow('canvasOutput', dst);
dst.delete();
});
canvas.addEventListener('dispose', e => {
src.delete();
tool.delete();
});
</script>
<script type="text/javascript">
let utils = new Utils('errorMessage');
utils.loadCode('codeSnippet', 'codeEditor');
utils.loadImageToCanvas('lena.jpg', 'canvasInput');
utils.addFileInputHandler('fileInput', 'canvasInput');
let disposeEvent = new Event('dispose');
let tryIt = document.getElementById('tryIt');
let stopIt = document.getElementById('stopIt');
tryIt.addEventListener('click', () => {
let e_input = document.getElementById('inputParams');
e_input.style.display = 'none';
let e_result = document.getElementById("result")
e_result.style.display = '';
var e = document.getElementById("canvasOutput");
var e_new = e.cloneNode(true);
e.parentNode.replaceChild(e_new, e); // reset event handlers
stopIt.removeAttribute('disabled');
tryIt.setAttribute('disabled', '');
utils.executeCode('codeEditor');
});
stopIt.addEventListener('click', () => {
let e_input = document.getElementById('inputParams');
e_input.style.display = '';
let e_result = document.getElementById("result")
e_result.style.display = 'none';
var e = document.getElementById("canvasOutput");
e.dispatchEvent(disposeEvent);
var e_new = e.cloneNode(true);
e.parentNode.replaceChild(e_new, e); // reset event handlers
tryIt.removeAttribute('disabled');
stopIt.setAttribute('disabled', '');
});
utils.loadOpenCv(() => {
tryIt.removeAttribute('disabled');
});
</script>
</body>
</html>

14
doc/js_tutorials/js_imgproc/js_intelligent_scissors/js_intelligent_scissors.markdown Normal file
View File

@ -0,0 +1,14 @@
Intelligent Scissors Demo {#tutorial_js_intelligent_scissors}
=========================
Goal
----
- Here you can check how to use IntelligentScissors tool for image segmentation task.
- Available methods and parameters: @ref cv::segmentation::IntelligentScissorsMB
\htmlonly
<iframe src="../../js_intelligent_scissors.html" width="100%"
onload="this.style.height=this.contentDocument.body.scrollHeight +'px';">
</iframe>
\endhtmlonly

4
doc/js_tutorials/js_imgproc/js_table_of_contents_imgproc.markdown
View File

@ -77,3 +77,7 @@ Image Processing {#tutorial_js_table_of_contents_imgproc}
- @subpage tutorial_js_imgproc_camera - @subpage tutorial_js_imgproc_camera
Learn image processing for video capture. Learn image processing for video capture.
- @subpage tutorial_js_intelligent_scissors
Learn how to use IntelligentScissors tool for image segmentation task.

7
doc/opencv.bib
View File

@ -768,6 +768,13 @@
pages = {432--441}, pages = {432--441},
publisher = {Springer} publisher = {Springer}
} }
@INPROCEEDINGS{Mortensen95intelligentscissors,
author = {Eric N. Mortensen and William A. Barrett},
title = {Intelligent Scissors for Image Composition},
booktitle = {In Computer Graphics, SIGGRAPH Proceedings},
year = {1995},
pages = {191--198}
}
@inproceedings{Muja2009, @inproceedings{Muja2009,
author = {Muja, Marius and Lowe, David G}, author = {Muja, Marius and Lowe, David G},
title = {Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration}, title = {Fast Approximate Nearest Neighbors with Automatic Algorithm Configuration},

19
modules/imgproc/include/opencv2/imgproc.hpp
View File

@ -185,6 +185,7 @@ location of points on the plane, building special graphs (such as NNG,RNG), and
@defgroup imgproc_motion Motion Analysis and Object Tracking @defgroup imgproc_motion Motion Analysis and Object Tracking
@defgroup imgproc_feature Feature Detection @defgroup imgproc_feature Feature Detection
@defgroup imgproc_object Object Detection @defgroup imgproc_object Object Detection
@defgroup imgproc_segmentation Image Segmentation
@defgroup imgproc_c C API @defgroup imgproc_c C API
@defgroup imgproc_hal Hardware Acceleration Layer @defgroup imgproc_hal Hardware Acceleration Layer
@{ @{
@ -3227,6 +3228,9 @@ CV_EXPORTS_AS(EMD) float wrapperEMD( InputArray signature1, InputArray signature
//! @} imgproc_hist //! @} imgproc_hist
//! @addtogroup imgproc_segmentation
//! @{
/** @example samples/cpp/watershed.cpp /** @example samples/cpp/watershed.cpp
An example using the watershed algorithm An example using the watershed algorithm
*/ */
@ -3254,11 +3258,11 @@ function.
size as image . size as image .
@sa findContours @sa findContours
@ingroup imgproc_misc
*/ */
CV_EXPORTS_W void watershed( InputArray image, InputOutputArray markers ); CV_EXPORTS_W void watershed( InputArray image, InputOutputArray markers );
//! @} imgproc_segmentation
//! @addtogroup imgproc_filter //! @addtogroup imgproc_filter
//! @{ //! @{
@ -3304,7 +3308,7 @@ CV_EXPORTS_W void pyrMeanShiftFiltering( InputArray src, OutputArray dst,
//! @} //! @}
//! @addtogroup imgproc_misc //! @addtogroup imgproc_segmentation
//! @{ //! @{
/** @example samples/cpp/grabcut.cpp /** @example samples/cpp/grabcut.cpp
@ -3334,6 +3338,11 @@ CV_EXPORTS_W void grabCut( InputArray img, InputOutputArray mask, Rect rect,
InputOutputArray bgdModel, InputOutputArray fgdModel, InputOutputArray bgdModel, InputOutputArray fgdModel,
int iterCount, int mode = GC_EVAL ); int iterCount, int mode = GC_EVAL );
//! @} imgproc_segmentation
//! @addtogroup imgproc_misc
//! @{
/** @example samples/cpp/distrans.cpp /** @example samples/cpp/distrans.cpp
An example on using the distance transform An example on using the distance transform
*/ */
@ -4876,4 +4885,8 @@ Point LineIterator::pos() const
} // cv } // cv
#include "./imgproc/segmentation.hpp"
#endif #endif

141
modules/imgproc/include/opencv2/imgproc/segmentation.hpp Normal file
View File

@ -0,0 +1,141 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#ifndef OPENCV_IMGPROC_SEGMENTATION_HPP
#define OPENCV_IMGPROC_SEGMENTATION_HPP
#include "opencv2/imgproc.hpp"
namespace cv {
namespace segmentation {
//! @addtogroup imgproc_segmentation
//! @{
/** @brief Intelligent Scissors image segmentation
*
* This class is used to find the path (contour) between two points
* which can be used for image segmentation.
*
* Usage example:
* @snippet snippets/imgproc_segmentation.cpp usage_example_intelligent_scissors
*
* Reference: <a href="http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.138.3811&rep=rep1&type=pdf">"Intelligent Scissors for Image Composition"</a>
* algorithm designed by Eric N. Mortensen and William A. Barrett, Brigham Young University
* @cite Mortensen95intelligentscissors
*/
class CV_EXPORTS_W_SIMPLE IntelligentScissorsMB
{
public:
CV_WRAP
IntelligentScissorsMB();
/** @brief Specify weights of feature functions
*
* Consider keeping weights normalized (sum of weights equals to 1.0)
* Discrete dynamic programming (DP) goal is minimization of costs between pixels.
*
* @param weight_non_edge Specify cost of non-edge pixels (default: 0.43f)
* @param weight_gradient_direction Specify cost of gradient direction function (default: 0.43f)
* @param weight_gradient_magnitude Specify cost of gradient magnitude function (default: 0.14f)
*/
CV_WRAP
IntelligentScissorsMB& setWeights(float weight_non_edge, float weight_gradient_direction, float weight_gradient_magnitude);
/** @brief Specify gradient magnitude max value threshold
*
* Zero limit value is used to disable gradient magnitude thresholding (default behavior, as described in original article).
* Otherwize pixels with `gradient magnitude >= threshold` have zero cost.
*
* @note Thresholding should be used for images with irregular regions (to avoid stuck on parameters from high-contract areas, like embedded logos).
*
* @param gradient_magnitude_threshold_max Specify gradient magnitude max value threshold (default: 0, disabled)
*/
CV_WRAP
IntelligentScissorsMB& setGradientMagnitudeMaxLimit(float gradient_magnitude_threshold_max = 0.0f);
/** @brief Switch to "Laplacian Zero-Crossing" edge feature extractor and specify its parameters
*
* This feature extractor is used by default according to article.
*
* Implementation has additional filtering for regions with low-amplitude noise.
* This filtering is enabled through parameter of minimal gradient amplitude (use some small value 4, 8, 16).
*
* @note Current implementation of this feature extractor is based on processing of grayscale images (color image is converted to grayscale image first).
*
* @note Canny edge detector is a bit slower, but provides better results (especially on color images): use setEdgeFeatureCannyParameters().
*
* @param gradient_magnitude_min_value Minimal gradient magnitude value for edge pixels (default: 0, check is disabled)
*/
CV_WRAP
IntelligentScissorsMB& setEdgeFeatureZeroCrossingParameters(float gradient_magnitude_min_value = 0.0f);
/** @brief Switch edge feature extractor to use Canny edge detector
*
* @note "Laplacian Zero-Crossing" feature extractor is used by default (following to original article)
*
* @sa Canny
*/
CV_WRAP
IntelligentScissorsMB& setEdgeFeatureCannyParameters(
double threshold1, double threshold2,
int apertureSize = 3, bool L2gradient = false
);
/** @brief Specify input image and extract image features
*
* @param image input image. Type is #CV_8UC1 / #CV_8UC3
*/
CV_WRAP
IntelligentScissorsMB& applyImage(InputArray image);
/** @brief Specify custom features of imput image
*
* Customized advanced variant of applyImage() call.
*
* @param non_edge Specify cost of non-edge pixels. Type is CV_8UC1. Expected values are `{0, 1}`.
* @param gradient_direction Specify gradient direction feature. Type is CV_32FC2. Values are expected to be normalized: `x^2 + y^2 == 1`
* @param gradient_magnitude Specify cost of gradient magnitude function: Type is CV_32FC1. Values should be in range `[0, 1]`.
* @param image **Optional parameter**. Must be specified if subset of features is specified (non-specified features are calculated internally)
*/
CV_WRAP
IntelligentScissorsMB& applyImageFeatures(
InputArray non_edge, InputArray gradient_direction, InputArray gradient_magnitude,
InputArray image = noArray()
);
/** @brief Prepares a map of optimal paths for the given source point on the image
*
* @note applyImage() / applyImageFeatures() must be called before this call
*
* @param sourcePt The source point used to find the paths
*/
CV_WRAP void buildMap(const Point& sourcePt);
/** @brief Extracts optimal contour for the given target point on the image
*
* @note buildMap() must be called before this call
*
* @param targetPt The target point
* @param[out] contour The list of pixels which contains optimal path between the source and the target points of the image. Type is CV_32SC2 (compatible with `std::vector<Point>`)
* @param backward Flag to indicate reverse order of retrived pixels (use "true" value to fetch points from the target to the source point)
*/
CV_WRAP void getContour(const Point& targetPt, OutputArray contour, bool backward = false) const;
#ifndef CV_DOXYGEN
struct Impl;
inline Impl* getImpl() const { return impl.get(); }
protected:
std::shared_ptr<Impl> impl;
#endif
};
//! @}
} // namespace segmentation
} // namespace cv
#endif // OPENCV_IMGPROC_SEGMENTATION_HPP

772
modules/imgproc/src/intelligent_scissors.cpp Normal file
View File

@ -0,0 +1,772 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
//
// Copyright (C) 2020, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
#include "precomp.hpp"
//#include "opencv2/imgproc/segmentation.hpp"
#include <opencv2/core/utils/logger.hpp>
#include <queue> // std::priority_queue
namespace cv {
namespace segmentation {
namespace {
// 0 1 2
// 3 x 4
// 5 6 7
static const int neighbors[8][2] = {
{ -1, -1 },
{ 0, -1 },
{ 1, -1 },
{ -1, 0 },
{ 1, 0 },
{ -1, 1 },
{ 0, 1 },
{ 1, 1 },
};
// encoded reverse direction
static const int neighbors_encode[8] = {
7+1, 6+1, 5+1,
4+1, 3+1,
2+1, 1+1, 0+1
};
#define ACOS_TABLE_SIZE 64
// acos_table[x + ACOS_TABLE_SIZE] = acos(x / ACOS_TABLE_SIZE) / CV_PI (see local_cost)
// x = [ -ACOS_TABLE_SIZE .. ACOS_TABLE_SIZE ]
float* getAcosTable()
{
constexpr int N = ACOS_TABLE_SIZE;
static bool initialized = false;
static float acos_table[2*N + 1] = { 0 };
if (!initialized)
{
const float CV_PI_inv = static_cast<float>(1.0 / CV_PI);
for (int i = -N; i <= N; i++)
{
acos_table[i + N] = acosf(i / (float)N) * CV_PI_inv;
}
initialized = true;
}
return acos_table;
}
} // namespace anon
struct IntelligentScissorsMB::Impl
{
// proposed weights from the article (sum = 1.0)
float weight_non_edge = 0.43f;
float weight_gradient_direction = 0.43f;
float weight_gradient_magnitude = 0.14f;
enum EdgeFeatureMode {
FEATURE_ZERO_CROSSING = 0,
FEATURE_CANNY
};
EdgeFeatureMode edge_mode = FEATURE_ZERO_CROSSING;
// FEATURE_ZERO_CROSSING
float edge_gradient_magnitude_min_value = 0.0f;
// FEATURE_CANNY
double edge_canny_threshold1 = 10;
double edge_canny_threshold2 = 100;
int edge_canny_apertureSize = 3;
bool edge_canny_L2gradient = false;
float gradient_magnitude_threshold_max = 0.0f; // disabled thresholding
int sobelKernelSize = 3; // 1 or 3
int laplacianKernelSize = 3; // 1 or 3
// image features
Mat_<Point2f> gradient_direction; //< I: normalized laplacian x/y components
Mat_<float> gradient_magnitude; //< Fg: gradient cost function
Mat_<uchar> non_edge_feature; //< Fz: zero-crossing function
float weight_non_edge_compute = 0.0f;
// encoded paths map (produced by `buildMap()`)
Mat_<uchar> optimalPathsMap;
void resetFeatures_()
{
CV_TRACE_FUNCTION();
gradient_direction.release();
gradient_magnitude.release();
non_edge_feature.release();
weight_non_edge_compute = weight_non_edge;
optimalPathsMap.release();
}
Size src_size;
Mat image_;
Mat grayscale_;
void initImage_(InputArray image)
{
CV_TRACE_FUNCTION();
if (!image_.empty())
return;
CV_CheckType(image.type(), image.type() == CV_8UC1 || image.type() == CV_8UC3 || image.type() == CV_8UC4, "");
src_size = image.size();
image_ = image.getMat();
}
void initGrayscale_(InputArray image)
{
CV_TRACE_FUNCTION();
if (!grayscale_.empty())
return;
CV_Assert(!image.empty());
CV_CheckType(image.type(), image.type() == CV_8UC1 || image.type() == CV_8UC3 || image.type() == CV_8UC4, "");
src_size = image.size();
if (image.channels() > 1)
cvtColor(image, grayscale_, COLOR_BGR2GRAY);
else
grayscale_ = image.getMat();
}
Mat Ix_, Iy_;
void initImageDerives_(InputArray image)
{
CV_TRACE_FUNCTION();
if (!Ix_.empty())
return;
initGrayscale_(image);
Sobel(grayscale_, Ix_, CV_32FC1, 1, 0, sobelKernelSize);
Sobel(grayscale_, Iy_, CV_32FC1, 0, 1, sobelKernelSize);
}
Mat image_magnitude_;
void initImageMagnitude_(InputArray image)
{
CV_TRACE_FUNCTION();
if (!image_magnitude_.empty())
return;
initImageDerives_(image);
magnitude(Ix_, Iy_, image_magnitude_);
}
void cleanupFeaturesTemporaryArrays_()
{
CV_TRACE_FUNCTION();
image_.release();
grayscale_.release();
Ix_.release();
Iy_.release();
image_magnitude_.release();
}
Impl()
{
// nothing
CV_TRACE_FUNCTION();
}
void setWeights(float weight_non_edge_, float weight_gradient_direction_, float weight_gradient_magnitude_)
{
CV_TRACE_FUNCTION();
CV_CheckGE(weight_non_edge_, 0.0f, "");
CV_CheckGE(weight_gradient_direction_, 0.0f, "");
CV_CheckGE(weight_gradient_magnitude_, 0.0f, "");
CV_CheckGE(weight_non_edge_ + weight_gradient_direction_ + weight_gradient_magnitude_, FLT_EPSILON, "Sum of weights must be greater than zero");
weight_non_edge = weight_non_edge_;
weight_gradient_direction = weight_gradient_direction_;
weight_gradient_magnitude = weight_gradient_magnitude_;
resetFeatures_();
}
void setGradientMagnitudeMaxLimit(float gradient_magnitude_threshold_max_)
{
CV_TRACE_FUNCTION();
CV_CheckGE(gradient_magnitude_threshold_max_, 0.0f, "");
gradient_magnitude_threshold_max = gradient_magnitude_threshold_max_;
resetFeatures_();
}
void setEdgeFeatureZeroCrossingParameters(float gradient_magnitude_min_value_)
{
CV_TRACE_FUNCTION();
CV_CheckGE(gradient_magnitude_min_value_, 0.0f, "");
edge_mode = FEATURE_ZERO_CROSSING;
edge_gradient_magnitude_min_value = gradient_magnitude_min_value_;
resetFeatures_();
}
void setEdgeFeatureCannyParameters(
double threshold1, double threshold2,
int apertureSize = 3, bool L2gradient = false
)
{
CV_TRACE_FUNCTION();
CV_CheckGE(threshold1, 0.0, "");
CV_CheckGE(threshold2, 0.0, "");
edge_mode = FEATURE_CANNY;
edge_canny_threshold1 = threshold1;
edge_canny_threshold2 = threshold2;
edge_canny_apertureSize = apertureSize;
edge_canny_L2gradient = L2gradient;
resetFeatures_();
}
void applyImageFeatures(
InputArray non_edge, InputArray gradient_direction_, InputArray gradient_magnitude_,
InputArray image
)
{
CV_TRACE_FUNCTION();
resetFeatures_();
cleanupFeaturesTemporaryArrays_();
src_size = Size(0, 0);
if (!non_edge.empty())
src_size = non_edge.size();
if (!gradient_direction_.empty())
{
Size gradient_direction_size = gradient_direction_.size();
if (!src_size.empty())
CV_CheckEQ(src_size, gradient_direction_size, "");
else
src_size = gradient_direction_size;
}
if (!gradient_magnitude_.empty())
{
Size gradient_magnitude_size = gradient_magnitude_.size();
if (!src_size.empty())
CV_CheckEQ(src_size, gradient_magnitude_size, "");
else
src_size = gradient_magnitude_size;
}
if (!image.empty())
{
Size image_size = image.size();
if (!src_size.empty())
CV_CheckEQ(src_size, image_size, "");
else
src_size = image_size;
}
// src_size must be filled
CV_Assert(!src_size.empty());
if (!non_edge.empty())
{
CV_CheckTypeEQ(non_edge.type(), CV_8UC1, "");
non_edge_feature = non_edge.getMat();
}
else
{
if (weight_non_edge == 0.0f)
{
non_edge_feature.create(src_size);
non_edge_feature.setTo(0);
}
else
{
if (image.empty())
CV_Error(Error::StsBadArg, "Non-edge feature parameter is missing. Input image parameter is required to extract this feature");
extractEdgeFeature_(image);
}
}
if (!gradient_direction_.empty())
{
CV_CheckTypeEQ(gradient_direction_.type(), CV_32FC2, "");
gradient_direction = gradient_direction_.getMat();
}
else
{
if (weight_gradient_direction == 0.0f)
{
gradient_direction.create(src_size);
gradient_direction.setTo(Scalar::all(0));
}
else
{
if (image.empty())
CV_Error(Error::StsBadArg, "Gradient direction feature parameter is missing. Input image parameter is required to extract this feature");
extractGradientDirection_(image);
}
}
if (!gradient_magnitude_.empty())
{
CV_CheckTypeEQ(gradient_magnitude_.type(), CV_32FC1, "");
gradient_magnitude = gradient_magnitude_.getMat();
}
else
{
if (weight_gradient_magnitude == 0.0f)
{
gradient_magnitude.create(src_size);
gradient_magnitude.setTo(Scalar::all(0));
}
else
{
if (image.empty())
CV_Error(Error::StsBadArg, "Gradient magnitude feature parameter is missing. Input image parameter is required to extract this feature");
extractGradientMagnitude_(image);
}
}
cleanupFeaturesTemporaryArrays_();
}
void extractEdgeFeature_(InputArray image)
{
CV_TRACE_FUNCTION();
if (edge_mode == FEATURE_CANNY)
{
CV_LOG_DEBUG(NULL, "Canny(" << edge_canny_threshold1 << ", " << edge_canny_threshold2 << ")");
Mat img_canny;
Canny(image, img_canny, edge_canny_threshold1, edge_canny_threshold2, edge_canny_apertureSize, edge_canny_L2gradient);
#if 0
threshold(img_canny, non_edge_feature, 254, 1, THRESH_BINARY_INV);
#else
// Canny result values are 0 or 255
bitwise_not(img_canny, non_edge_feature);
weight_non_edge_compute = weight_non_edge * (1.0f / 255.0f);
#endif
}
else // if (edge_mode == FEATURE_ZERO_CROSSING)
{
initGrayscale_(image);
Mat_<short> laplacian;
Laplacian(grayscale_, laplacian, CV_16S, laplacianKernelSize);
Mat_<uchar> zero_crossing(src_size, 1);
const size_t zstep = zero_crossing.step[0];
for (int y = 0; y < src_size.height - 1; y++)
{
const short* row0 = laplacian.ptr<short>(y);
const short* row1 = laplacian.ptr<short>(y + 1);
uchar* zrow0 = zero_crossing.ptr<uchar>(y);
//uchar* zrow1 = zero_crossing.ptr<uchar>(y + 1);
for (int x = 0; x < src_size.width - 1; x++)
{
const int v = row0[x];
const int neg_v = -v;
// - * 1
// 2 3 4
const int v1 = row0[x + 1];
const int v2 = (x > 0) ? row1[x - 1] : v;
const int v3 = row1[x + 0];
const int v4 = row1[x + 1];
if (v < 0)
{
if (v1 > 0)
{
zrow0[x + ((v1 < neg_v) ? 1 : 0)] = 0;
}
if (v2 > 0)
{
zrow0[x + ((v2 < neg_v) ? (zstep - 1) : 0)] = 0;
}
if (v3 > 0)
{
zrow0[x + ((v3 < neg_v) ? (zstep + 0) : 0)] = 0;
}
if (v4 > 0)
{
zrow0[x + ((v4 < neg_v) ? (zstep + 1) : 0)] = 0;
}
}
else
{
if (v1 < 0)
{
zrow0[x + ((v1 > neg_v) ? 1 : 0)] = 0;
}
if (v2 < 0)
{
zrow0[x + ((v2 > neg_v) ? (zstep - 1) : 0)] = 0;
}
if (v3 < 0)
{
zrow0[x + ((v3 > neg_v) ? (zstep + 0) : 0)] = 0;
}
if (v4 < 0)
{
zrow0[x + ((v4 > neg_v) ? (zstep + 1) : 0)] = 0;
}
}
}
}
if (edge_gradient_magnitude_min_value > 0)
{
initImageMagnitude_(image);
Mat mask = image_magnitude_ < edge_gradient_magnitude_min_value;
zero_crossing.setTo(1, mask); // reset low-amplitude noise
}
non_edge_feature = zero_crossing;
}
}
void extractGradientDirection_(InputArray image)
{
CV_TRACE_FUNCTION();
initImageMagnitude_(image); // calls internally: initImageDerives_(image);
gradient_direction.create(src_size);
for (int y = 0; y < src_size.height; y++)
{
const float* magnutude_row = image_magnitude_.ptr<float>(y);
const float* Ix_row = Ix_.ptr<float>(y);
const float* Iy_row = Iy_.ptr<float>(y);
Point2f* gradient_direction_row = gradient_direction.ptr<Point2f>(y);
for (int x = 0; x < src_size.width; x++)
{
const float m = magnutude_row[x];
if (m > FLT_EPSILON)
{
float m_inv = 1.0f / m;
gradient_direction_row[x] = Point2f(Ix_row[x] * m_inv, Iy_row[x] * m_inv);
}
else
{
gradient_direction_row[x] = Point2f(0, 0);
}
}
}
}
void extractGradientMagnitude_(InputArray image)
{
CV_TRACE_FUNCTION();
initImageMagnitude_(image); // calls internally: initImageDerives_(image);
Mat m;
double max_m = 0;
if (gradient_magnitude_threshold_max > 0)
{
threshold(image_magnitude_, m, gradient_magnitude_threshold_max, 0, THRESH_TRUNC);
max_m = gradient_magnitude_threshold_max;
}
else
{
m = image_magnitude_;
minMaxLoc(m, 0, &max_m);
}
if (max_m <= FLT_EPSILON)
{
CV_LOG_INFO(NULL, "IntelligentScissorsMB: input image gradient is almost zero")
gradient_magnitude.create(src_size);
gradient_magnitude.setTo(0);
}
else
{
m.convertTo(gradient_magnitude, CV_32F, -1.0 / max_m, 1.0); // normalize and inverse to range 0..1
}
}
void applyImage(InputArray image)
{
CV_TRACE_FUNCTION();
CV_CheckType(image.type(), image.type() == CV_8UC1 || image.type() == CV_8UC3 || image.type() == CV_8UC4, "");
resetFeatures_();
cleanupFeaturesTemporaryArrays_();
extractEdgeFeature_(image);
extractGradientDirection_(image);
extractGradientMagnitude_(image);
cleanupFeaturesTemporaryArrays_();
}
// details: see section 3.1 of the article
const float* acos_table = getAcosTable();
float local_cost(const Point& p, const Point& q) const
{
const bool isDiag = (p.x != q.x) && (p.y != q.y);
float fG = gradient_magnitude.at<float>(q);
const Point2f diff((float)(q.x - p.x), (float)(q.y - p.y));
const Point2f Ip = gradient_direction(p);
const Point2f Iq = gradient_direction(q);
const Point2f Dp(Ip.y, -Ip.x); // D(p) - 90 degrees clockwise
const Point2f Dq(Iq.y, -Iq.x); // D(q) - 90 degrees clockwise
float dp = Dp.dot(diff); // dp(p, q)
float dq = Dq.dot(diff); // dq(p, q)
if (dp < 0)
{
dp = -dp; // ensure dp >= 0
dq = -dq;
}
const float sqrt2_inv = 0.7071067811865475f; // 1.0 / sqrt(2)
if (isDiag)
{
dp *= sqrt2_inv; // normalize length of (q - p)
dq *= sqrt2_inv; // normalize length of (q - p)
}
else
{
fG *= sqrt2_inv;
}
#if 1
int dp_i = cvFloor(dp * ACOS_TABLE_SIZE); // dp is in range 0..1
dp_i = std::min(ACOS_TABLE_SIZE, std::max(0, dp_i));
int dq_i = cvFloor(dq * ACOS_TABLE_SIZE); // dq is in range -1..1
dq_i = std::min(ACOS_TABLE_SIZE, std::max(-ACOS_TABLE_SIZE, dq_i));
const float fD = acos_table[dp_i + ACOS_TABLE_SIZE] + acos_table[dq_i + ACOS_TABLE_SIZE];
#else
const float CV_PI_inv = static_cast<float>(1.0 / CV_PI);
const float fD = (acosf(dp) + acosf(dq)) * CV_PI_inv; // TODO optimize acos calls (through tables)
#endif
float cost =
weight_non_edge_compute * non_edge_feature.at<uchar>(q) +
weight_gradient_direction * fD +
weight_gradient_magnitude * fG;
return cost;
}
struct Pix
{
Point pt;
float cost; // NOTE: do not remove cost from here through replacing by cost(pt) map access
inline bool operator > (const Pix &b) const
{
return cost > b.cost;
}
};
void buildMap(const Point& start_point)
{
CV_TRACE_FUNCTION();
CV_Assert(!src_size.empty());
CV_Assert(!gradient_magnitude.empty() && "Features are missing. applyImage() must be called first");
CV_CheckGE(weight_non_edge + weight_gradient_direction + weight_gradient_magnitude, FLT_EPSILON, "");
#if 0 // debug
Rect wholeImage(0, 0, src_size.width, src_size.height);
Rect roi = Rect(start_point.x - 5, start_point.y - 5, 11, 11) & wholeImage;
std::cout << roi << std::endl;
std::cout << gradient_magnitude(roi) << std::endl;
std::cout << gradient_direction(roi) << std::endl;
std::cout << non_edge_feature(roi) << std::endl;
#endif
optimalPathsMap.release();
optimalPathsMap.create(src_size);
optimalPathsMap.setTo(0); // optimalPathsMap(start_point) = 0;
//
// Section 3.2
// Live-Wire 2-D DP graph search.
//
Mat_<float> cost_map(src_size, FLT_MAX); // g(q)
Mat_<uchar> processed(src_size, (uchar)0); // e(q)
// Note: std::vector is faster than std::deque
// TODO check std::set
std::priority_queue< Pix, std::vector<Pix>, std::greater<Pix> > L;
cost_map(start_point) = 0;
L.emplace(Pix{ start_point, 0/*cost*/ });
while (!L.empty())
{
Pix pix = L.top(); L.pop();
Point q = pix.pt; // 'q' from the article
if (processed(q))
continue; // already processed (with lower cost, see note below)
processed(q) = 1;
#if 1
const float cost_q = pix.cost;
#else
const float cost_q = cost_map(q);
CV_Assert(cost_q == pix.cost);
#endif
for (int n = 0; n < 8; n++) // scan neighbours
{
Point r(q.x + neighbors[n][0], q.y + neighbors[n][1]); // 'r' from the article
if (r.x < 0 || r.x >= src_size.width || r.y < 0 || r.y >= src_size.height)
continue; // out of range
#if !defined(__EMSCRIPTEN__) // slower in JS
float& cost_r = cost_map(r);
if (cost_r < cost_q)
continue; // already processed
#else
if (processed(r))
continue; // already processed
float& cost_r = cost_map(r);
CV_DbgCheckLE(cost_q, cost_r, "INTERNAL ERROR: sorted queue is corrupted");
#endif
float cost = cost_q + local_cost(q, r); // TODO(opt): compute partially until cost < cost_r
if (cost < cost_r)
{
#if 0 // avoid compiler warning
if (cost_r != FLT_MAX)
{
// In article the point 'r' is removed from the queue L
// to be re-inserted again with sorting against new optimized cost.
// We can do nothing, because "new point" will be placed before in the sorted queue.
// Old point will be skipped through "if (processed(q))" check above after processing of new optimal candidate.
//
// This approach leads to some performance impact, however it is much smaller than element removal from the sorted queue.
// So, do nothing.
}
#endif
cost_r = cost;
L.emplace(Pix{ r, cost });
optimalPathsMap(r) = (uchar)neighbors_encode[n];
}
}
}
}
void getContour(const Point& target, OutputArray contour_, bool backward)
{
CV_TRACE_FUNCTION();
CV_Assert(!optimalPathsMap.empty() && "buildMap() must be called before getContour()");
const int cols = optimalPathsMap.cols;
const int rows = optimalPathsMap.rows;
std::vector<Point> result; result.reserve(512);
size_t loop_check = 4096;
Point pt = target;
for (size_t i = 0; i < (size_t)rows * cols; i++) // don't hang on invalid maps
{
CV_CheckLT(pt.x, cols, "");
CV_CheckLT(pt.y, rows, "");
result.push_back(pt);
int direction = (int)optimalPathsMap(pt);
if (direction == 0)
break; // stop, start point is reached
CV_CheckLT(direction, 9, "Map is invalid");
Point next(pt.x + neighbors[direction - 1][0], pt.y + neighbors[direction - 1][1]);
pt = next;
if (result.size() == loop_check) // optional sanity check of invalid maps with loops (don't eat huge amount of memory)
{
loop_check *= 4; // next limit for loop check
for (const auto& pt_check : result)
{
CV_CheckNE(pt_check, pt, "Map is invalid. Contour loop is detected");
}
}
}
if (backward)
{
_InputArray(result).copyTo(contour_);
}
else
{
const int N = (int)result.size();
const int sz[1] = { N };
contour_.create(1, sz, CV_32SC2);
Mat_<Point> contour = contour_.getMat();
for (int i = 0; i < N; i++)
{
contour.at<Point>(i) = result[N - (i + 1)];
}
}
}
};
IntelligentScissorsMB::IntelligentScissorsMB()
: impl(std::make_shared<Impl>())
{
// nothing
}
IntelligentScissorsMB& IntelligentScissorsMB::setWeights(float weight_non_edge, float weight_gradient_direction, float weight_gradient_magnitude)
{
CV_DbgAssert(impl);
impl->setWeights(weight_non_edge, weight_gradient_direction, weight_gradient_magnitude);
return *this;
}
IntelligentScissorsMB& IntelligentScissorsMB::setGradientMagnitudeMaxLimit(float gradient_magnitude_threshold_max)
{
CV_DbgAssert(impl);
impl->setGradientMagnitudeMaxLimit(gradient_magnitude_threshold_max);
return *this;
}
IntelligentScissorsMB& IntelligentScissorsMB::setEdgeFeatureZeroCrossingParameters(float gradient_magnitude_min_value)
{
CV_DbgAssert(impl);
impl->setEdgeFeatureZeroCrossingParameters(gradient_magnitude_min_value);
return *this;
}
IntelligentScissorsMB& IntelligentScissorsMB::setEdgeFeatureCannyParameters(
double threshold1, double threshold2,
int apertureSize, bool L2gradient
)
{
CV_DbgAssert(impl);
impl->setEdgeFeatureCannyParameters(threshold1, threshold2, apertureSize, L2gradient);
return *this;
}
IntelligentScissorsMB& IntelligentScissorsMB::applyImage(InputArray image)
{
CV_DbgAssert(impl);
impl->applyImage(image);
return *this;
}
IntelligentScissorsMB& IntelligentScissorsMB::applyImageFeatures(
InputArray non_edge, InputArray gradient_direction, InputArray gradient_magnitude,
InputArray image
)
{
CV_DbgAssert(impl);
impl->applyImageFeatures(non_edge, gradient_direction, gradient_magnitude, image);
return *this;
}
void IntelligentScissorsMB::buildMap(const Point& pt)
{
CV_DbgAssert(impl);
impl->buildMap(pt);
}
void IntelligentScissorsMB::getContour(const Point& target, OutputArray contour, bool backward) const
{
CV_DbgAssert(impl);
impl->getContour(target, contour, backward);
}
}} // namespace

467
modules/imgproc/test/test_intelligent_scissors.cpp Normal file
View File

@ -0,0 +1,467 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
#include "test_precomp.hpp"
//#include "opencv2/imgproc/segmentation.hpp"
namespace opencv_test { namespace {
Mat getTestImageGray()
{
static Mat m;
if (m.empty())
{
m = imread(findDataFile("shared/lena.png"), IMREAD_GRAYSCALE);
}
return m.clone();
}
Mat getTestImageColor()
{
static Mat m;
if (m.empty())
{
m = imread(findDataFile("shared/lena.png"), IMREAD_COLOR);
}
return m.clone();
}
Mat getTestImage1()
{
static Mat m;
if (m.empty())
{
m.create(Size(200, 100), CV_8UC1);
m.setTo(Scalar::all(128));
Rect roi(50, 30, 100, 40);
m(roi).setTo(Scalar::all(0));
#if 0
imshow("image", m);
waitKey();
#endif
}
return m.clone();
}
Mat getTestImage2()
{
static Mat m;
if (m.empty())
{
m.create(Size(200, 100), CV_8UC1);
m.setTo(Scalar::all(128));
Rect roi(40, 30, 100, 40);
m(roi).setTo(Scalar::all(255));
#if 0
imshow("image", m);
waitKey();
#endif
}
return m.clone();
}
Mat getTestImage3()
{
static Mat m;
if (m.empty())
{
m.create(Size(200, 100), CV_8UC1);
m.setTo(Scalar::all(128));
Scalar color(0,0,0,0);
line(m, Point(30, 50), Point(50, 50), color, 1);
line(m, Point(50, 50), Point(80, 30), color, 1);
line(m, Point(150, 50), Point(80, 30), color, 1);
line(m, Point(150, 50), Point(180, 50), color, 1);
line(m, Point(80, 10), Point(80, 90), Scalar::all(200), 1);
line(m, Point(100, 10), Point(100, 90), Scalar::all(200), 1);
line(m, Point(120, 10), Point(120, 90), Scalar::all(200), 1);
#if 0
imshow("image", m);
waitKey();
#endif
}
return m.clone();
}
Mat getTestImage4()
{
static Mat m;
if (m.empty())
{
m.create(Size(200, 100), CV_8UC1);
for (int y = 0; y < m.rows; y++)
{
for (int x = 0; x < m.cols; x++)
{
float dx = (float)(x - 100);
float dy = (float)(y - 100);
float d = sqrtf(dx * dx + dy * dy);
m.at<uchar>(y, x) = saturate_cast<uchar>(100 + 100 * sin(d / 10 * CV_PI));
}
}
#if 0
imshow("image", m);
waitKey();
#endif
}
return m.clone();
}
Mat getTestImage5()
{
static Mat m;
if (m.empty())
{
m.create(Size(200, 100), CV_8UC1);
for (int y = 0; y < m.rows; y++)
{
for (int x = 0; x < m.cols; x++)
{
float dx = (float)(x - 100);
float dy = (float)(y - 100);
float d = sqrtf(dx * dx + dy * dy);
m.at<uchar>(y, x) = saturate_cast<uchar>(x / 2 + 100 * sin(d / 10 * CV_PI));
}
}
#if 0
imshow("image", m);
waitKey();
#endif
}
return m.clone();
}
void show(const Mat& img, const std::vector<Point> pts)
{
if (cvtest::debugLevel >= 10)
{
Mat dst = img.clone();
std::vector< std::vector<Point> > contours;
contours.push_back(pts);
polylines(dst, contours, false, Scalar::all(255));
imshow("dst", dst);
waitKey();
}
}
TEST(Imgproc_IntelligentScissorsMB, rect)
{
segmentation::IntelligentScissorsMB tool;
tool.applyImage(getTestImage1());
Point source_point(50, 30);
tool.buildMap(source_point);
Point target_point(100, 30);
std::vector<Point> pts;
tool.getContour(target_point, pts);
tool.applyImage(getTestImage2());
tool.buildMap(source_point);
std::vector<Point> pts2;
tool.getContour(target_point, pts2, true/*backward*/);
EXPECT_EQ(pts.size(), pts2.size());
}
TEST(Imgproc_IntelligentScissorsMB, lines)
{
segmentation::IntelligentScissorsMB tool;
Mat image = getTestImage3();
tool.applyImage(image);
Point source_point(30, 50);
tool.buildMap(source_point);
Point target_point(150, 50);
std::vector<Point> pts;
tool.getContour(target_point, pts);
EXPECT_EQ((size_t)121, pts.size());
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, circles)
{
segmentation::IntelligentScissorsMB tool;
tool.setGradientMagnitudeMaxLimit(10);
Mat image = getTestImage4();
tool.applyImage(image);
Point source_point(50, 50);
tool.buildMap(source_point);
Point target_point(150, 50);
std::vector<Point> pts;
tool.getContour(target_point, pts);
EXPECT_EQ((size_t)101, pts.size());
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, circles_gradient)
{
segmentation::IntelligentScissorsMB tool;
Mat image = getTestImage5();
tool.applyImage(image);
Point source_point(50, 50);
tool.buildMap(source_point);
Point target_point(150, 50);
std::vector<Point> pts;
tool.getContour(target_point, pts);
EXPECT_EQ((size_t)101, pts.size());
show(image, pts);
}
#define PTS_SIZE_EPS 2
TEST(Imgproc_IntelligentScissorsMB, grayscale)
{
segmentation::IntelligentScissorsMB tool;
Mat image = getTestImageGray();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 206;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, check_features_grayscale_1_0_0_zerro_crossing_with_limit)
{
segmentation::IntelligentScissorsMB tool;
tool.setEdgeFeatureZeroCrossingParameters(64);
tool.setWeights(1.0f, 0.0f, 0.0f);
Mat image = getTestImageGray();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 207;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, check_features_grayscale_1_0_0_canny)
{
segmentation::IntelligentScissorsMB tool;
tool.setEdgeFeatureCannyParameters(50, 100);
tool.setWeights(1.0f, 0.0f, 0.0f);
Mat image = getTestImageGray();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 201;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, check_features_grayscale_0_1_0)
{
segmentation::IntelligentScissorsMB tool;
tool.setWeights(0.0f, 1.0f, 0.0f);
Mat image = getTestImageGray();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 166;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, check_features_grayscale_0_0_1)
{
segmentation::IntelligentScissorsMB tool;
tool.setWeights(0.0f, 0.0f, 1.0f);
Mat image = getTestImageGray();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 197;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, color)
{
segmentation::IntelligentScissorsMB tool;
Mat image = getTestImageColor();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 205;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, color_canny)
{
segmentation::IntelligentScissorsMB tool;
tool.setEdgeFeatureCannyParameters(32, 100);
Mat image = getTestImageColor();
tool.applyImage(image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 200;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, color_custom_features_invalid)
{
segmentation::IntelligentScissorsMB tool;
ASSERT_ANY_THROW(tool.applyImageFeatures(noArray(), noArray(), noArray()));
}
TEST(Imgproc_IntelligentScissorsMB, color_custom_features_edge)
{
segmentation::IntelligentScissorsMB tool;
Mat image = getTestImageColor();
Mat canny_edges;
Canny(image, canny_edges, 32, 100, 5);
Mat binary_edge_feature;
cv::threshold(canny_edges, binary_edge_feature, 254, 1, THRESH_BINARY_INV);
tool.applyImageFeatures(binary_edge_feature, noArray(), noArray(), image);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 201;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, color_custom_features_all)
{
segmentation::IntelligentScissorsMB tool;
tool.setWeights(0.9f, 0.0f, 0.1f);
Mat image = getTestImageColor();
Mat canny_edges;
Canny(image, canny_edges, 50, 100, 5);
Mat binary_edge_feature; // 0, 1 values
cv::threshold(canny_edges, binary_edge_feature, 254, 1, THRESH_BINARY_INV);
Mat_<Point2f> gradient_direction(image.size(), Point2f(0, 0)); // normalized
Mat_<float> gradient_magnitude(image.size(), 0); // cost function
tool.applyImageFeatures(binary_edge_feature, gradient_direction, gradient_magnitude);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 201;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
TEST(Imgproc_IntelligentScissorsMB, color_custom_features_edge_magnitude)
{
segmentation::IntelligentScissorsMB tool;
tool.setWeights(0.9f, 0.0f, 0.1f);
Mat image = getTestImageColor();
Mat canny_edges;
Canny(image, canny_edges, 50, 100, 5);
Mat binary_edge_feature; // 0, 1 values
cv::threshold(canny_edges, binary_edge_feature, 254, 1, THRESH_BINARY_INV);
Mat_<float> gradient_magnitude(image.size(), 0); // cost function
tool.applyImageFeatures(binary_edge_feature, noArray(), gradient_magnitude);
Point source_point(275, 63);
tool.buildMap(source_point);
Point target_point(413, 155);
std::vector<Point> pts;
tool.getContour(target_point, pts);
size_t gold = 201;
EXPECT_GE(pts.size(), gold - PTS_SIZE_EPS);
EXPECT_LE(pts.size(), gold + PTS_SIZE_EPS);
show(image, pts);
}
}} // namespace

2
modules/js/src/core_bindings.cpp
View File

@ -87,6 +87,8 @@ namespace hal {
using namespace emscripten; using namespace emscripten;
using namespace cv; using namespace cv;
using namespace cv::segmentation; // FIXIT
#ifdef HAVE_OPENCV_DNN #ifdef HAVE_OPENCV_DNN
using namespace cv::dnn; using namespace cv::dnn;
#endif #endif

23
modules/js/test/test_imgproc.js
View File

@ -977,3 +977,26 @@ QUnit.test('warpPolar', function(assert) {
96, 83, 64, 45, 32 96, 83, 64, 45, 32
]); ]);
}); });
QUnit.test('IntelligentScissorsMB', function(assert) {
const lines = new cv.Mat(50, 100, cv.CV_8U, new cv.Scalar(0));
lines.row(10).setTo(new cv.Scalar(255));
assert.ok(lines instanceof cv.Mat);
let tool = new cv.segmentation_IntelligentScissorsMB();
tool.applyImage(lines);
assert.ok(lines instanceof cv.Mat);
lines.delete();
tool.buildMap(new cv.Point(10, 10));
let contour = new cv.Mat();
tool.getContour(new cv.Point(50, 10), contour);
assert.equal(contour.type(), cv.CV_32SC2);
assert.ok(contour.total() == 41, contour.total());
tool.getContour(new cv.Point(80, 10), contour);
assert.equal(contour.type(), cv.CV_32SC2);
assert.ok(contour.total() == 71, contour.total());
});

17
platforms/js/opencv_js.config.py
View File

@ -18,8 +18,21 @@ imgproc = {'': ['Canny', 'GaussianBlur', 'Laplacian', 'HoughLines', 'HoughLinesP
'matchShapes', 'matchTemplate','medianBlur', 'minAreaRect', 'minEnclosingCircle', 'moments', 'morphologyEx', \ 'matchShapes', 'matchTemplate','medianBlur', 'minAreaRect', 'minEnclosingCircle', 'moments', 'morphologyEx', \
'pointPolygonTest', 'putText','pyrDown','pyrUp','rectangle','remap', 'resize','sepFilter2D','threshold', \ 'pointPolygonTest', 'putText','pyrDown','pyrUp','rectangle','remap', 'resize','sepFilter2D','threshold', \
'undistort','warpAffine','warpPerspective','warpPolar','watershed', \ 'undistort','warpAffine','warpPerspective','warpPolar','watershed', \
'fillPoly', 'fillConvexPoly'], 'fillPoly', 'fillConvexPoly', 'polylines',
'CLAHE': ['apply', 'collectGarbage', 'getClipLimit', 'getTilesGridSize', 'setClipLimit', 'setTilesGridSize']} ],
'CLAHE': ['apply', 'collectGarbage', 'getClipLimit', 'getTilesGridSize', 'setClipLimit', 'setTilesGridSize'],
'segmentation_IntelligentScissorsMB': [
'IntelligentScissorsMB',
'setWeights',
'setGradientMagnitudeMaxLimit',
'setEdgeFeatureZeroCrossingParameters',
'setEdgeFeatureCannyParameters',
'applyImage',
'applyImageFeatures',
'buildMap',
'getContour'
],
}
objdetect = {'': ['groupRectangles'], objdetect = {'': ['groupRectangles'],
'HOGDescriptor': ['load', 'HOGDescriptor', 'getDefaultPeopleDetector', 'getDaimlerPeopleDetector', 'setSVMDetector', 'detectMultiScale'], 'HOGDescriptor': ['load', 'HOGDescriptor', 'getDefaultPeopleDetector', 'getDaimlerPeopleDetector', 'setSVMDetector', 'detectMultiScale'],

13
samples/cpp/CMakeLists.txt
View File

@ -23,6 +23,9 @@ if(NOT BUILD_EXAMPLES OR NOT OCV_DEPENDENCIES_FOUND)
return() return()
endif() endif()
set(DEPS_example_snippet_imgproc_segmentation opencv_core opencv_imgproc)
set(DEPS_example_cpp_intelligent_scissors opencv_core opencv_imgproc opencv_imgcodecs opencv_highgui)
project(cpp_samples) project(cpp_samples)
ocv_include_modules_recurse(${OPENCV_CPP_SAMPLES_REQUIRED_DEPS}) ocv_include_modules_recurse(${OPENCV_CPP_SAMPLES_REQUIRED_DEPS})
file(GLOB_RECURSE cpp_samples RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.cpp) file(GLOB_RECURSE cpp_samples RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.cpp)
@ -32,11 +35,17 @@ endif()
ocv_list_filterout(cpp_samples "real_time_pose_estimation/") ocv_list_filterout(cpp_samples "real_time_pose_estimation/")
foreach(sample_filename ${cpp_samples}) foreach(sample_filename ${cpp_samples})
set(package "cpp") set(package "cpp")
if(sample_filename MATCHES "tutorial_code") if(sample_filename MATCHES "tutorial_code/snippet")
set(package "snippet")
elseif(sample_filename MATCHES "tutorial_code")
set(package "tutorial") set(package "tutorial")
endif() endif()
ocv_define_sample(tgt ${sample_filename} ${package}) ocv_define_sample(tgt ${sample_filename} ${package})
ocv_target_link_libraries(${tgt} PRIVATE ${OPENCV_LINKER_LIBS} ${OPENCV_CPP_SAMPLES_REQUIRED_DEPS}) set(deps ${OPENCV_CPP_SAMPLES_REQUIRED_DEPS})
if(DEFINED DEPS_${tgt})
set(deps ${DEPS_${tgt}})
endif()
ocv_target_link_libraries(${tgt} PRIVATE ${OPENCV_LINKER_LIBS} ${deps})
if(sample_filename MATCHES "/gpu/" AND HAVE_opencv_cudaarithm AND HAVE_opencv_cuda_filters) if(sample_filename MATCHES "/gpu/" AND HAVE_opencv_cudaarithm AND HAVE_opencv_cuda_filters)
ocv_target_link_libraries(${tgt} PRIVATE opencv_cudaarithm opencv_cudafilters) ocv_target_link_libraries(${tgt} PRIVATE opencv_cudaarithm opencv_cudafilters)
endif() endif()

35
samples/cpp/tutorial_code/snippets/imgproc_segmentation.cpp Normal file
View File

@ -0,0 +1,35 @@
#include "opencv2/imgproc.hpp"
#include "opencv2/imgproc/segmentation.hpp"
using namespace cv;
static
void usage_example_intelligent_scissors()
{
Mat image(Size(1920, 1080), CV_8UC3, Scalar::all(128));
//! [usage_example_intelligent_scissors]
segmentation::IntelligentScissorsMB tool;
tool.setEdgeFeatureCannyParameters(16, 100) // using Canny() as edge feature extractor
.setGradientMagnitudeMaxLimit(200);
// calculate image features
tool.applyImage(image);
// calculate map for specified source point
Point source_point(200, 100);
tool.buildMap(source_point);
// fast fetching of contours
// for specified target point and the pre-calculated map (stored internally)
Point target_point(400, 300);
std::vector<Point> pts;
tool.getContour(target_point, pts);
//! [usage_example_intelligent_scissors]
}
int main()
{
usage_example_intelligent_scissors();
return 0;
}

3
samples/samples_utils.cmake
View File

@ -4,6 +4,9 @@
function(ocv_define_sample out_target source sub) function(ocv_define_sample out_target source sub)
get_filename_component(name "${source}" NAME_WE) get_filename_component(name "${source}" NAME_WE)
set(the_target "example_${sub}_${name}") set(the_target "example_${sub}_${name}")
if(OPENCV_DUMP_EXAMPLE_TARGET)
message(STATUS "Example: ${the_target} (${source})")
endif()
add_executable(${the_target} "${source}") add_executable(${the_target} "${source}")
if(TARGET Threads::Threads AND NOT OPENCV_EXAMPLES_DISABLE_THREADS) if(TARGET Threads::Threads AND NOT OPENCV_EXAMPLES_DISABLE_THREADS)
target_link_libraries(${the_target} PRIVATE Threads::Threads) target_link_libraries(${the_target} PRIVATE Threads::Threads)