DCT-based edge detector for snapshot images

In this paper, we propose a discrete cosine transform (DCT)-based edge detector for snapshot images that are obtained by performing decimation (by a factor of two or more) on a much larger original image. We first partition the original image into non-overlapping blocks, where the block size equals the decimation factor, and then perform DCT on each block. The variances of the DCT's AC coefficients are used to decide the edge map of snapshot images. A directional edge map of the snapshot images can also be generated if the variance of the directional AC coefficients in the vertical, horizontal, or diagonal direction is used. Simulation results show that the proposed DCT-based edge detector performs better (based on an objective PSNR criterion) than the conventional technique - applying an edge detector on the down-sampled image.