Shulou(Shulou.com)06/03 Report--

This article will explain in detail why the edge detection effect of opencv's canny function is different from that of matlab, and the content of the article is of high quality, so the editor will share it for you to do a reference. I hope you will have a certain understanding of the relevant knowledge after reading this article.

It is obvious that the edge of matlab is more delicate.

First, review the main steps of the traditional canny algorithm:

1. The x and y directional derivatives of grayscale image are obtained by using sobel difference operator.

2. Find out the gradient size and direction of each point in the image.

3. Set two thresholds of high and low. If the gradient is greater than the high threshold, it is a strong edge pixel, and if it is greater than the low threshold, it may be a weak edge point.

4. Non-maximum suppression is carried out along the gradient direction in the strong edge points left after a screening.

5. Find the adjacent weak edge points along the strong edge points after the second screening to get the final edge.

Opencv and matlab basically follow the above steps to get their own canny functions, but generally speaking, the image is blurred and smoothed before processing, so the effect is better. Opencv pays more attention to real-time, so even smoothing is put outside the function as an option, while matlab pays more attention to quality. In addition to adding smoothing operation, there are a series of other optimization operations.

1. Fuzzy smoothing

In fact, this term has a great impact on the generated edge effect, the smoother the smoother, the smoother the resulting edge, the same use of Gaussian low-pass filtering Then the size of the filter and the variance of the Gaussian distribution are two key parameters. Determine what value is the best through experiments, but generally use 5 × 5 and variance 2. Opencv's cvSmooth function can be smoothed, generally using an image whose input and output are 8 bits deep, while matlab converts the image into a floating-point type for smoothing, so matlab is better in accuracy. Because this directly determines the accuracy of the two directional derivatives obtained by the later derivation operation.

2. Gradient

In the application of sobel, the two are consistent, but the algorithm for finding the direction of the gradient is completely different, but in essence, the effect is basically the same after the author's test. In terms of gradient, opencv provides two measurement methods of Euclidean distance and Hamiltonian distance. The former is the sum of squares and the latter is the sum of direct absolute values. Opencv uses the latter by default. We know that ((x2+y2) / 2) 1Compact 2 > = (| x | + | y |) / 2, and | x | + | y | > = (x2+y2) 1p2, so theoretically using Hamiltonian distance is basically consistent with the real value, but there is always a difference in accuracy. This is the second reason why some people say that opencv's canny is not as good as matlab. In cv.h, you can switch by modifying CV_CANNY_L2_GRADIENT.

3. Threshold

The threshold here is whether the gradient can be passed by the switch, opencv thresholds are manually set, while matlab is to make a histogram of the strength of all points, retaining a certain percentage of strong edges, the difference in this step will not affect the quality of the results, but of course there will be differences in speed, which is the contradiction between speed and convenience.

4. Thin edge effect

The edge after non-maximum suppression is basically the width of one or two pixels, and finally hopes to get a linewidth edge. Opencv only uses a simple judgment to prevent two strong edge pixels from appearing in parallel or juxtaposed. This simple thin edge technique will lead to edge fracture. Matlab uses Louisa Lam, Seong-Whan Lee, and Ching Y. Wuen, "Thinning Methodologies-A Comprehensive Survey," IEEE TrPAMI, vol. 14, no. 9, pp. 869-885, 1992. The method of this article carries on the thin edge treatment twice, the effect is quite good, and the edge is very smooth.

To sum up, if the effect of opencv is similar to that of matlab, efforts must be made on smoothing, the size of the gradient should be measured by Euclidean distance, and the thin edge algorithm of matlab should be added, so that the processing time will probably double, but it can also meet the real-time processing. In addition, in general visual applications, the algorithm can be modified to a threshold to remove the step of strong and weak edge connection to improve speed.

About why opencv's canny function edge detection effect and matlab is different to share here, I hope the above content can be of some help to you, can learn more knowledge. If you think the article is good, you can share it for more people to see.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.