Color quantification method based on density peak value

Abstract

The invention discloses a color quantification method based on a density peak value, and the method comprises the steps: letting an image comprise N pixel points, calculating the Euclidean distances dij between each two pixel points, and building a distance matrix D; letting a cut-off distance be dc, and calculating a local density rho (i) of each pixel point according to the distance matrix D; for each pixel point, finding all pixel points which have higher local densities than the pixel point; finding the Euclidean distances from the pixel point to the pixel points with higher local densities according to the distance matrix D, and defining a minimum Euclidean distance delta (i); calculating a parameter gamma (i), equal to rho (i) * delta (i), of each pixel point, enabling all parameters gamma (i) to be arranged in a descending order, and selecting the pixel points corresponding to the front K parameters gamma (i) as cluster center points; enabling each pixel point after the K pixel points to be ranked in types corresponding to the cluster center points, and then employing a mean value of each type to represent the pixel value of this type, thereby completing the color quantification. The method is good in performance, can effectively and robustly achieve color quantification, and can flexibly meet different needs.

Description

technical field [0001] The invention relates to a color quantization method. In particular, it relates to a color quantization method based on density peaks. Background technique [0002] Color quantization (color quantization, CQ) is a method to reduce the color types of digital color images while maintaining a good visual effect of the reconstructed image. There are tens of millions of colors in true-color images, which brings great difficulties to their storage, transmission, display, and processing. For example, for a medium-resolution true-color image (resolution 640×480, 24bit / pixel), it can display more than 16 million colors, and the number of bits is 640×480×24=7.37Mb, which takes About 0.9MB of storage space. Reducing the color types through color quantization can reduce the number of encoding bits of the image, which is beneficial to the storage, transmission and display of the image. Besides that, color quantization also reduces the feature space of an image,...

AI Technical Summary

Problems solved by technology

Compared with forward clustering, backward clustering is better quantified because they involve loop iterations or stochastic optimization steps, but at the same time the calculation time is increased

Images