Image retrieval based on color distribution entropy

Morphological Description of Color Images for Content-Based Image Retrieval

“…Color is widely regarded as one of the most expressive visual features [20], and as such it has been extensively studied in the context of CBIR, thus leading to a rich variety of descriptors [21]. Similarly to texture and shape, it is relatively hard to establish the exact description of color content.…”

“…As they are capable of providing the first two elements required for the description of color, their foremost drawback is the lack of any spatial information. To this end, several extensions have been proposed [35], [21]; among which one particular approach is using multiple resolutions, that has proven itself to be able to capture effectively color content [36], [37]. Multiresolution decomposition has been achieved by means of Gaussian filters [36], Gabor filters [38] as well as wavelets [39].…”

“…In detail, since highly simplified images are bound to be more alike than the lower resolutions where more image detail is preserved, it makes sense to weight the corresponding histogram intersections. For instance, given and , the multiresolution histogram families for two images and , respectively, their weighted distance would become (21) where denotes the th histogram bin of resolution . Further refinement may be achieved by modifying the bin width at each resolution.…”

“…In our tests, the proposed morphological color descriptors have been compared against the standard histogram, the autocorrelogram (AC) [31], the superiority of which with respect to the histogram has been shown by multiple comparative studies [22], [25], the color structure descriptor (CSD) [20] of the MPEG-7 standard and color distribution entropy (CDE) [21].…”

“…But the question is, how exactly should one quantize this space? Common models include 4-4-16 [20], 3-3-12 [25], as well as 3-3-9 [21]. Moreover one has also the option of quantizing the hue circle nonuniformly, according to [52], which is based on the fact that human color perception has a varying capacity of hue distinction.…”