Comparison of Color Space Models Based on Human Color Classification and Image Segmentation

In t,his paper we compared color space models l,ased on hrunan color perception. For this purpose, we made t.wo experinlents to see how human suh-ject,s classify color pixels in color ima.ges. In the first. experiment, subject,^ classify color pixels without seeing the ima.ge. In the second experiment, the subjects segment the images int.0 regions considered t,o I' e of similar colors, seeing the images. We define three crit,eria, two of which are based on the color classification and segmentation made by humans , for comparison of eleven color space models. As a conclusion, we found t,hat L*u*v* and L*a*b* color moclels are most closer t,o human color percep-In color image segmentation based on clustering met.hotls, segment,at.ion resu1t.s vary according t.o color spa.ce model used. There is, however, no guideline for selecting t,he color space moclel. In our earlier work[l] we compa.red eight color moilels fro111 t,he viewpoint, of using then1 in t,he ISO-DATA clustering method for color image segmenta-t.ion. However, the problenl of choosing an appropriate color nod el arises not only in segmentmation of images. By generalizing the problem, we would like to find color space(s) close to human color percep-t,ion. \Ve compa.re eleven color space nlodels ba.sed on t.he result,s of pixel color classificat,ion and region segment.at.ion performed by hulllan subjects.