The most distinct character of mosaic objects in digital images is that their boundaries are usually presented in grid or semi-grid shape. Therefore the first task for searching mosaic objects is to identifying these grids or semi-grid blocks in digital images. Canny edge detector is employed to improve performance in identifying edge points, which are thus used to search mosaic points by template matching. To ensure the efficiency of edge detection, the optimal threshold for pixels classification is studied in this paper. At first, the initial thresholds are obtained by an iterative method, then 1/2, 1/4, 1/8, and 1/16 of the initial lower one is used as lower threshold in searching mosaic points. Fifty images are selected as experimental subjects, and applied with different thresholds. The results show when lower threshold is set to 1/4 of the initial, mosaic points can be detected, but with a low hit rate; when set to 1/8 of the original, mosaic points can be identified with acceptable accuracy.