Color quantization is the process of finding an acceptable set of colors that can be used to represent the original colors of a digital image. It is a lossy compression technique that transforms a set containing a large number of colors to a set that contains a relatively small number of colors. The ultimate goal of any color-quantizing algorithm is to minimize the perceptual differences between the original and quantized images.

Quantization of color digital images offers a number of significant advantages. First and foremost is the fact that color-quantized images can be displayed on low-cost graphics systems which are normally incapable of showing full-color images. Another advantage is that less bandwidth is required to transmit a color-quantized image. Finally, images that are color quantized require considerably less storage space than their full-color counterparts.

In this presentation, a new algorithm that color quantizes video sequences is described. It creates multiple color palettes for a video sequence yet avoids the extreme case of creating a different palette for each frame. It exploits the fact that there are many instances during a video sequence where minor color changes occur between consecutive frames. The proposed algorithm performs clustering by grouping the colors of a frame into partitions using a method based on principal components. Prior to clustering, Hotelling's T-squared statistic is used to determine whether or not the colors in the current frame are significantly different than those of the frame where clustering last occurred. If the algorithm detects a significant difference, it performs a clustering phase to calculate a new color palette. The algorithm skips the clustering phase if a significant difference is not detected. Quantized frames are generated by mapping each color in the original frame to its nearest neighbor in the color palette.

The new algorithm produces high-quality quantized video sequences by addressing the weaknesses of currently used techniques. Instead of computing a new color palette for each frame or requiring that a single color palette suffice for an entire video sequence, the proposed algorithm creates multiple palettes for a video sequence but computes new a color palette only when necessary.