Generally, feature extraction is highly subjective in the sense that what is the problem you are trying to solve. However, the feature selection could be to some extent objective, where you try to select subset of good features from large feature set.
The usual way to deal with images is to design "Filters" based on what features you want to extract. These are used all the time in image processing and re-touching software such as Photoshop. The usual way is that the image is decomposed using some transform ( FFT , SVD, PCA , Wavelet etc.. ) then for example in FFT, if you want to extract the low frequency information from the image you design a "low pass" filter which only extracts low frequency information and leaves the high frequencies, this gives a blurring effect. Similarly, a "high pass" filter is used for edge detection ( sharpening the image). So based on what you want to do with the image (what features you want) you design different filters.