The described technology is generally directed towards comparing two images for content differences, such as images based on the frames of a show and a derivative version of that show. Frame pairs such as from an original show and its derivative version are processed into image pairs, which can include decoding, scaling, luminance extraction and/or filtering. An edge pixel image is obtained via edge detection for each image. Edge pixels in one image are compared against a counterpart edge pixel (and possibly neighboring pixels) in the other image to determine matches (matching edge pixels) and mismatches. An image with too many errors based on the mismatches is deemed as a candidate for further review. A difference image can be generated to assist a reviewer in detecting where the mismatches were detected. By repeating for the various frames, a show can be automatically compared against its derivative for content differences.