Study of MSE and SSIM of Canny Edge Detection

Traditionally for using the Canny edge detection algorithm, direct use of library is made. By this we directly get the output image. But what is observed is that the mean squared error is generally high. So in this paper we have implemented the algorithm without using the library. The result obtained is the number of intermediate images of smoothning, gradient ,non maximum suppression and threshold images and a final output image after performing hysteresis. We see that the mean squared error of base image and output image that we get without using the library is less compared to the mean squared error of base image and output image that we got using the library. To summaries everything we can say that when we do not make use of the library we are actually changing the threshold value and noting down the results at different threshold value. It is known that CDA is very sensitive to noise so before detecting any edges first we have used gaussian filter and convolution. After removing the noise sobel filter is used for finding the gradients. After that thresholding is used to identify weak and strong edges then by using hysteresis method decision is made whether to remove or to keep the weak edges. After all this steps are completed we can observe that the mse of ground truth image and output image is significantly reduced. Whereas in case where library is used,we import the libraries numpy, cv2 and matplotlib then read the input image on grayscale. After that simply apply the library Canny on the grayscale image to get the output image. The Structural similarity index is calculated between ground truth image and gradient, non maximum suppression , threshold images to see how much they are similar. The sum of the difference between the two images is calculated and observed that it was higher. Keywords - Structural Similarity Index, Mean Squared Error, Canny Edge Detection, Sobel filter