Chaos Image Encryption

In the past two days, I have been looking at the data of chaotic image encryption, and I have thought of a lot of methods, but in many cases, they are doing operations at the pixel level. I don't think this is a good direction. For example, there are only 256 cases in a grayscale image, and when I use matlab to do operations, I can't directly operate on the pixel values. Only after the pixel value is converted to a double type can it be used for operations. I don't think this is very good, because the pixel value loses numerical precision when converting to double, which in some subtle cases can have an impact on the file. Although it does not affect the use, it can be regarded as a hidden danger in terms of safety. So I want to directly convert the pixel value to binary, this situation can be avoided.