Key points are not available for this paper at this time.
The question of using the mean squared error (MSE) measure as a quality criterion in image processing is addressed. The MSE was evaluated as a measure of perceived similarity between an original and processed picture by comparing computed MSE values with similarity judgements. In addition, several other measures that weighted the error in accordance with assumed properties of the visual system were studied. The results obtained indicated that MSE is not very valid as a quality criterion for pictures reproduced for human viewing and the improved measures could be derived by weighting the error in accordance with assumed properties of the visual system. However, none of the tested measures were valid for all kinds of pictures. Thus picture properties also had to be incorporated in subjective error measures.
Hans Marmolin (Wed,) studied this question.