Which hash algorithm has the lowest collision probability for very similar photo/video files?

Question

Hashing

The storage contains 50 million photos and 50 million videos. The hash sum of each photo and video is needed in order to guarantee the immutability of the data. By collisions, I mean not the coincidence of the hash sums of different photos or videos, but a random change in data on the HDD (for example, a pixel in the photo has fallen off), in which the hash sum has not changed (there is such a theoretical probability). What hash algorithm gives the minimum probability of such a collision?
ps. The cryptographic strength of the algorithm is not needed. Only the minimum probability of collision is needed for different degrees of data similarity (photo and video).
pps It is not the number of unique hash values ​​of a given algorithm that matters. It is important that for any image file, firstly, there are as few collisions as possible per each bit of the output value of this hash algorithm, and secondly, that the distribution of collisions over the entire space of theoretically possible images for it should be as large as possible. uniform, and did not have a pronounced "dome" with the center from this file.