基于贝尔模板的文本图像压缩及重构算法的研究

The Raw Data Compression and Reconstruction of Text Image Based on Bayer CFA Pattern

针对文本图像提出了直接对CFA原始数据进行压缩的算法,即首先把贝尔CFA图像从RGB色空间转换到YCrCb色空间,然后把不利于压缩的高频分量中比较多的亮度分量进行低通滤波,再用结构分离法把梅花形的亮度分量转化为矩形,最后分别采用JPEG2000压缩亮度和色度通道。为了估计解压恢复的CFA模板丢失的彩色分量,应用改进的自适应模糊匹配插值算法,对它周围四个相邻连接的像素分配一个模糊隶属度,从而利用相邻信息,且把这个隶属度作为加权平均工具来重构全彩图像。通过实验验证,这种贝尔数据直接压缩方法带来的失真很小,插值后恢复的全彩图像在主观视觉和客观质量评价中也能获得令人满意的效果。

This paper introduced a CFA raw data direct compression algorithm for text image. This algorithm firstly converted color space from RGB to YCrCb for Bayer CFA image, then performed low-pass filter for luminance component which was not suitable for compression, and transformed shape construction of the luminance component pattern from quincunx to rectangle using structure separation method, finally compressed respectively to luminance component and chrominance component channel using JPEG2000. This paper also based on our need and applied an improved adaptive fuzzy matching based CFA interpolation algorithm in order to get color image for the decompressed recovered CFA pattern, which assigned a fuzzy membership to the surrounding four connected neighboring pixels and use this fuzzy membership assignment as a weighted-average tool to reconstruct the color image. Through many experiments, this direct compression method acquired lower distortion, and interpolated color image could produce excellent results in terms of both subjective and objective image quality measures.