A color-complexity feature is proposed to describe the color variations of the pixels of an image, and a color-spatial feature to state the pixel color distribution on different locations in an image. These two features complement each other, and mainly try to provide tolerance to shifting and noise. The authors have implemented an image retrieval system using these features separately and in a combined manner. Experimental results, based on 420 images, show that all three approaches provide effective results. Among the three, the combined approach is the best. The authors present performance data, and show that their results compare favorably with those of another study.

The simplicity of the approach is appealing. For example, the approach can be used in information retrieval (IR) classes to show the applicability of fundamental retrieval concepts, such as indexing and query-object matching, in image environments. The results are preliminary and inconclusive, however. The paper omits any discussion of issues such as scalability, namely the time and space complexity of the approach in real size applications. Previous experience shows that large-scale IR system implementations usually require enhancement of the data structures and indexing features of their prototypes.