how to find mean of an image

It depends what your image is of.

Suppose you have an image of a single cell, but it happens to be high resolution. Then, after you have removed the parts of the image that are outside of the boundaries of the single cell, you would perhaps be wanting to take the mean of the entire image.

But suppose instead you have an image of multiple cells, where only some of the cells were abnormal. You could still mask out areas that are outside any cell. If you take the mean of the part only inside the cells, then you could be taking the mean of a bunch of good cells with a small number of abnormal cells -- the more good cells you had the less the influence of the abnormal cells would be, so you could end up missing the abnormal cells. In such a case you would want to consider the mean of each cell separately.

Now, for any one cell, is it the case that the color is affected for the entire cell? Or is it the case that part of the cell is affected in color? If only part is affected in color, then the mean color over the entire cell is going to be affected by the proportion of affected area to non-affected area. And perhaps that is reasonable for your situation. But if you need to identify the part of the cell that is affected then you need to look at only portions of the cell at a time.

If you are attempting to look at portions of a cell at a time, then there is the question of how large an area you should examine. That can depend on what you are looking for. Ask yourself: if the image were at (say) 5x further magnification, then would you want the area you take of the mean of to grow with the magnification, or would you instead want the higher resolution to identify the abnormal section more precisely ?