Improved census stereo matching algorithm based on window cross-correlation information
A stereo matching and cross-correlation technology, applied in computing, computer components, image analysis, etc., can solve problems such as noise suppression defects, stereo matching algorithm stability discount, etc., and achieve high matching accuracy
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0073] Fig. 3 (a) and Fig. 3 (b) are four groups of standard reference color stereo images of Tsukuba, Teddy, Venus, Cones provided by the authoritative stereo matching algorithm testing platform Middlebury website to it, the present invention uses these four images to match Algorithms are evaluated. The sizes of the four image pairs are 384×288, 450×375, 434×383, and 450×375 pixels, respectively, and the left and right images have been calibrated and corrected. The first to fourth columns are respectively: the left image, the right image, the standard disparity map, the result map of the present invention, and the mismatch point label map.
[0074] It can be seen intuitively from Fig. 3(c) and Fig. 3(d) that the disparity map optimized by the present invention is relatively close to the standard disparity map as a whole, but there are still obvious differences in local details. After marking the mismatching points, it can be clearly seen that the four groups of mismatching p...
Embodiment 2
[0076] For DP (dynamic programming) Fig. 4 (c), AdaptWeight (adaptive weight) Fig. 4 (d), traditional Census Fig. 4 (e), SAD-Census Fig. 4 (f) and the present invention Fig. 4 (g) five The disparity maps generated by the two stereo matching algorithms are compared intuitively. It can be clearly seen that the DP algorithm has a greater disadvantage than other algorithms, and it is easy to cause image distortion and serious loss of edge information. Although the AdaptWeight algorithm has better edge details, there are many parallax holes in the disparity map, and it is necessary to use nearby reliable disparity values to fill the holes later. As a kind of non-parametric transformation, the traditional Census transformation has improved, but the overall effect is still poor. The SAD-Census algorithm further overcomes the noise interference and has richer edge information, but there are more mismatching points in the areas where the image depth changes greatly or the texture in...
Embodiment 3
[0078] By calculating the false matching rates of four test image pairs in Middlebury in different regions, and comparing them with the false matching rates of four stereo matching algorithms: DP, AdaptWeight, traditional Census and SAD-Census. As shown in Table 1 below, Nocc is the false matching rate of non-occluded areas, All is the false matching rate of all areas, Disc is the false matching rate of depth discontinuous areas, and Average represents the average false matching rate. The false match rate refers to the ratio of the calculated parallax value to the real pixel whose error is greater than the parallax tolerance in the entire image, which can be expressed as:
[0079]
[0080] In the above formula, N represents the total number of pixels in the disparity map; g x (x,y) and g s (x, y) represent the disparity value obtained by the test algorithm and the real disparity value respectively, δ thresh Indicates the parallax tolerance, the value is 1.
[0081] It ca...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


