algorithm - Computer Science Theory: Image Similarity -

so i'm trying run comparison of different images , wondering if point me in right direction basic metrics can take group of images.

assuming have 2 images, , b, pretty want data possible each can later programmatically compare them. things "general color", "general shape", etc. great.

if can me find specific properties , algorithms compute them great!

thanks!

edit: end goal here able have computer tell me how "similar" pictures are. if 2 images same in 1 blurred out face; should register similar. if 2 pictures different, computer should able tell.

what talking way general , non-specific.

image information formalised entropy.

what seem looking feature extraction , comparing these features. there tons of features can extracted lot of them irrelevant depending on differences in pictures.

there space domain , frequency domain descriptors of image each can useful here. can name more 100 descriptors in case, 1 sufficient or none useful.

pre-processing important, perhaps turn images grey-scale , compare them.

this field immensely diverse, need bit more specific.

(update)

what looking topic of hundreds if not thousands of scientific articles. well, perhaps simplistic approach can work.

so assuming question here not identifying objects , there no transform, translation, scale or rotation involved , dealing 2 images same 1 have more noise added upon it:

1) image domain (space domain): compare pixels 1 one , add square of differences. normalise value width*height - divide number of pixels. useful measure of similarity.

2) frequency domain: convert image frequency domain image (using ftt in image processing tool such opencv) 2d well. same above squared diff above, perhaps want limit frequencies. normalise number of pixels. fares better on noise , translation , on small rotation not on scale.