權重MinMax彩色影像量化法之改進

Abstract

就影像處理而言, 彩色影像量化是一個產生高品質壓縮影像非常有用的技巧。在這篇論文中, 我們提出一個新的彩色影像量化技巧, 改良自權重MinMax 演算法。我們的方法是MinMax 和K-Means 這兩種演算法的混合並基於權重的histogram而產生的新的演算法。我們的演算法並不嚐試去尋求量化問題的最佳解答, 取而代之的是去尋求每一個群最佳的切割方式, 最主要的觀念是基於將每一個群內部的最大群距最小化。不同於權重MinMax演算法的是我們的方法具有較佳的效率, 而且經過量化處理之後的影像品質也非常好。同時我們提出的改良式演算法亦具有容易實作的優點。

For image processing, the color image quantization is a useful technique to produce high quality compression image. In this thesis, we present a new color image quantization technique that modifies the weighted MinMax algorithm. It is a hybrid method of MinMax and K-Means algorithms and based on weighted histogram. Instead of trying to solve an optimization problem, our algorithm is aiming to get the best-spread representation of the cluster in the color space. The main concept is based on minimizing the maximum intercluster distance. Unlike weighted MinMax algorithm, our method is more efficient and the quality of quantized image is good. In the mean time it is easy to implement also.