the simplest solution is subtraction, and select a range to consider some acceptable differences or use histogram matching   

At initial level you can use template matching or structural similarity index matching but these are not reliable to use if you are using photos with different orientation, you have to detect features and then use classification algorithms to classify based on the extracted features. Principle Component Analysis is the easiest way to compare the photos based on eigenvectors and eigenvalues

Hi..

If you want to compare the images on the bases of visual similarity then extract the features from the images. The image features can be color feature, texture and shape based features. In literature a large number of models exists to represent the image using visual features. So first you have to identify the prominent features that are representing your image. Then you can make a comparison on the basis of those features.

My target is compare two photo for detecting sign language. Kind of OCR problem. I take a photo and compare with my database then output in nearest result.

hope this well help you