The invention relates to a non-uniform fingerprint extraction and comparison method for camera light response based on combined feature representation. Firstly, the input image is preprocessed such asnormalization, color space conversion and the like. Secondly, saturation, chrominance, brightness, edge region, texture region and flat region of the input image are detected as candidate regions ofinterest. Then, the image is cut into color blocks of specified size according to the above information, and these color blocks are divided into six subsets; Finally, the color blocks in six subsets are sent to the residual network respectively, and the non-uniform fingerprint of camera light response in each color block is extracted by depth learning method, and a fingerprint extraction system isformed by pre-training, which is used for source camera identification. The method has the advantages of uniqueness and good robustness, and can detect the non-uniformity fingerprint of camera lightresponse quickly and accurately, and is suitable for forensic identification, check anti-counterfeiting, image matching, information security and other fields.