A Modified Algorithm for Perceived Contrast Measure in Digital Images

In this paper, we propose an algorithm for the measure of local and global contrast in digital images. It applies locally, at various sub-sampled levels, a simplified computation of local contrast based on DOG and finally it recombines all the values to obtain a global measure. The proposed method comes from the modification of a previous algorithm with a different local measure of contrast and with a parameterized way to recombine color channels. As new approach we propose the idea of recombining the channels following measures taken from the image itself. Preliminary tests and results are presented and discussed. Introduction Since the first studies on contrast in images, it has emerged how arduous it could be to give a definition of contrast and, moreover, how subjective and related to the observation task or observer experience this definition could turn out to be. For this reason, the first approaches to this topic have confined themselves to study the phenomenon from rather limited points of view, operating in controlled situations and under very restrictive conditions, the so-called “void conditions”. After this very first experiments more complex measures have been devised, but a measure of contrast in images is still not clearly defined. Several measures have been proposed so far [1-5]. The classic approaches consist of global measures and for this reason they result inadequate in most of the cases. In fact, the study of contrast in an image at a global level provides only a measure related on the maximum global difference in lightness and in some cases chromaticity. The response of the human visual system depends much less on the absolute luminance than on the relation of its local variations. A very first measure of global contrast, in the case of sinusoids or other periodic patterns of symmetrical deviations ranging from max L to min L , is Michelson contrast [1]: min max min max L L L L C M + − = Michelson’s definition is not suitable for natural images because one or two points of extreme brightness or darkness can determine the contrast of the whole image. To overcome the limits of global measures, alternative measures have been developed in the 90’s, among them Tadmor and Tolhurst [3]. Tadmor and Tolhurst’s analysis of contrast [3] is based on the D.O.G. (Difference Of Gaussian) model, adapted to natural images. In the conventional model, the spatial sensitivity in the center of receptive-fields (central component) is described by a bi-dimensional Gaussian: r y r x ) Center(x,y