Convolution-based edge detection for image/video in block DCT domain

This paper presents a scheme for performing convolution operation directly on compressed images without decompressing them first. The use of such a scheme is demonstrated and discussed by showing the implementation of the Laplacian-of-Gaussian operator for edge detection. We present a complete evaluation of the different parameters involved in this process and show edge detection results on several real images through our proposed scheme. In each case, it is shown that the proposed scheme of directly performing convolution on the compressed data leads to not only a significant computation speedup but also yields better edges.