前言

东莞，晴，29至27度。忙了一天，终于可以写写东西了。今天继续昨天的话题，我们在昨天的例了基础上完善，通过匹配关键点求出映射从而找到场景中的已知对象。

目标

本文你将学习

1. 采用nodeitk的findHomography和perspectiveTransform进行对象识别。
2. 此外，例子基本包含nodeitk的一些基本数据结构的使用：NodeOpenCVMat, NodeOpenCVKeyPoint, NodeOpenCVPoint
3. 上述基本的数据结构在nodeitk版本稳定后将会在使用手册中说明

var node_itk = require('./node-itk');
var img_object = node_itk.cv.imread( "./images/box.png", node_itk.cv.CV_LOAD_IMAGE_GRAYSCALE );
var img_scene = node_itk.cv.imread( "./images/box_in_scene.png", node_itk.cv.CV_LOAD_IMAGE_GRAYSCALE );
minHessian = 400
detector = new node_itk.cv.NodeOpenCVFeatureDetector("SURF")
detector.Set("hessianThreshold", minHessian)
keypoints_object = detector.Detect( img_object );
keypoints_scene = detector.Detect( img_scene );
extractor = new node_itk.cv.NodeOpenCVDescriptorExtractor("SURF");
descriptors_object = extractor.Compute(img_object, keypoints_object)
descriptors_scene = extractor.Compute(img_scene, keypoints_scene)
matcher = new node_itk.cv.NodeOpenCVDescriptorMatcher("FlannBased");
matches = matcher.Match(descriptors_object, descriptors_scene);
max_dist=0
min_dist=100
for (var i = 0; i < descriptors_object.Rows(); i++ ) {
	dist = matches[i].GetDistance();
	if (dist < min_dist) min_dist = dist;
	if (dist > max_dist) max_dist = dist;
};
console.log("-- Max dist : " + max_dist + "\n")
console.log("-- Min dist : " + min_dist + "\n")
var good_matches = [];
for( var i = 0; i < descriptors_object.Rows(); i++ ){ 
	if( matches[i].GetDistance() <= 3*min_dist )
	{ good_matches.push( matches[i] ); }
}
img_matches = node_itk.cv.DrawMatches(img_object, keypoints_object, img_scene, keypoints_scene, good_matches);
var obj=[], scene=[];
for (var i = 0; i < good_matches.length; i++) {
	obj.push( keypoints_object[good_matches[i].GetQueryIdx()].PT() )
	scene.push( keypoints_scene[good_matches[i].GetTrainIdx()].PT() )
};

H = node_itk.cv.FindHomography( obj, scene, node_itk.cv.CV_RANSAC );

obj_corners = []
obj_corners[0] = new node_itk.cv.NodeOpenCVPoint("Point2d", [0,0])
obj_corners[1] = new node_itk.cv.NodeOpenCVPoint("Point2d", [img_object.Cols(),0])
obj_corners[2] = new node_itk.cv.NodeOpenCVPoint("Point2d", [img_object.Cols(),img_object.Rows()])
obj_corners[3] = new node_itk.cv.NodeOpenCVPoint("Point2d", [0,img_object.Rows()])

tmp = new node_itk.cv.NodeOpenCVPoint("Point2d", [img_object.Cols(),0]);
color = new node_itk.cv.NodeOpenCVScalar("Scalar", [0,255,0]);
scene_corners = node_itk.cv.PerspectiveTransform(obj_corners, H.res);
node_itk.cv.Line(img_matches, scene_corners[0].Add(tmp), scene_corners[1].Add(tmp), color, 2)
node_itk.cv.Line(img_matches, scene_corners[1].Add(tmp), scene_corners[2].Add(tmp), color, 2)
node_itk.cv.Line(img_matches, scene_corners[2].Add(tmp), scene_corners[3].Add(tmp), color, 2)
node_itk.cv.Line(img_matches, scene_corners[3].Add(tmp), scene_corners[0].Add(tmp), color, 2)
node_itk.cv.NamedWindow( "Good Matches & Object detection", node_itk.cv.CV_WINDOW_AUTOSIZE );
node_itk.cv.imshow( "Good Matches & Object detection", img_matches );
node_itk.cv.WaitKey ( 0 );

小结

本文是昨天话题的深化，代码依然比较简洁。这是nodeitk遵循的原则：以简单的方式快速实现图像处理应用。喜欢的朋友就点踩，想说点东西的就评论吧！^_^ 待续