This is a compression problem, not a security problem. As other answers point out, it's not a different problem from redundant text, and modern crypto is secure even in that case.
If file size is a problem, and the input is an uncompressed image with raw RGB or YUV data, compress first with lossless WebP, Jpeg2000, HEIF (an HEVC aka h.265 I-frame), or even PNG (although that compresses each row separately so you could have similarity between rows.) https://en.wikipedia.org/wiki/Lossless_compression#Raster_graphics
Modern image codecs will find redundancy and compress it; their final output is typically a bitstream from a compressor like Huffman, gzip/Deflate, or something that solves the same problem, and will hopefully have found redundancy between similar blocks across the image. A simple image will get much smaller, closer to its Kolmogorov complexiy.
Of course, image compression is much slower than just throwing the input bits through AES, so it's a speed/size tradeoff, although for highly-redundant images with simple compression it can still save a lot without being too slow.
"Decrypting" then involves decompressing, which is fast. Any metadata from the original image needs to be sent, too, to reconstruct the original file if you want this to be fully transparent. That's possible; lossless archive compression software like RAR and maybe 7-zip already have filters to handle image files specially.
Compressing before encrypting has the same problem as usual, revealing information about compressibility. This can be a specific problem in text that's always the same except for one small secret, so I guess that might be something to be careful of for images, too, in the rare use-case where the image data is of a form or something and the way it's filled in is with a fixed font or by shading boxes.
Is it better to encrypt before compression or vice versa? mentions some of the problems for text, with things like TLS. And that lossy compression to a constant bitrate for video or audio would avoid leaking info via metadata (size).
Lossy compression can achieve quite good perceptual quality (for human vision or hearing) while shrinking files a lot, especially for video (redundancy between frames is usually high) but also individual still images. If you're building a whole system for transmitting multimedia stuff securely, you could make lossy compression part of it. If you're just encrypting data transparently, that implies lossless.