Color Image Quantization Scheme Using DBSCAN with K-Means Algorithm

Color image quantization (CIQ) is one of the important and well-accepted application areas in the field of data compression where a truly colored image is mainly represented by less number of selected significant color pixels. CIQ is performed in two major phases, i.e., color palette design and pixel mapping. The performance of any CIQ depends on the construction of a proper color palette, and this construction process is computationally expensive. In this paper, we have proposed a color palette design algorithm where we have incorporated two different types of clustering algorithms like density-based spatial clustering of applications with noise (DBSCAN) and K-means. Initially, we have decomposed the color image into several non-overlapping blocks, and subsequently, we have employed DBSCAN on each block. This process has concerned for some sort of initial screening of representative color pixels. Further, we have obtained the desired size of color palette, employing K-means on the earlier selected representative color pixels. We have tested the proposed scheme on a set of benchmark test images and obtained the satisfactory results in terms of the visual quality of the reconstructed images. In case of designing the color palette, the proposed scheme requires less computational time compare with the conventional K-means algorithm.