基于贝尔模板像的插值算法和压缩算法及其应用研究.pdf

太原理工大学 博士学位论文 基于贝尔模板像的插值算法和压缩算法及其应用研究 姓名：程永强 申请学位级别：博士 专业：@ 指导教师：谢克明;张刚 @ 太原理工大学博士研究生学位论文 I 基于贝尔模板像的插值算法和压缩算法及其应用研究 摘摘 要要 近年来，随着贝尔模板像传感器在各类数字成像设备中的广泛应用， 与之相关的像处理技术及应用迅速发展成为一个重要的研究领域目前， 彩像重构算法和贝尔像编码算法是该领域内的重点研究方向为降 低成像前端的复杂性，本文在贝尔像的时频分析基础上，研究了彩 像重构算法和贝尔像压缩算法之间的联系，提出了新的贝尔像的编码 方法及其相应的像重构算法，并对部分成果进行开发应用论文的主要 创新工作如下： 1、以国际标准 JPEG-LS 的框架为基础，提出了基于因果插值的贝尔模 板像无损编码器，对预测器进行了改进，给出两种可选择的方案，具有 较好的去相关性能为提高编码增益，重新优化了上下文量化区间所设 计的编码器能按照光栅顺序逐点地实现对贝尔模板像编码，在低复杂度 和压缩率之间取得了较好的平衡该算法可作为自然灰度像国际编码标 准 JPEG-LS 的扩充部分。

2、为进一步提高贝尔模板像压缩率，提出了基于小波子带替换的贝 尔像视觉无损压缩方法采用预测差分法无损压缩贝尔像绿分量， 用于有效地保护贝尔像原始亮度信息，并应用简单的线性插值方法估计 红蓝分量位置的绿值，分别对红蓝分量及其同位置的绿估计分量进行 小波变换， 进而得到红绿、 蓝绿低频子带差信号， 通过低复杂度的 JPEG-LS 太原理工大学博士研究生学位论文 II 进行无损或近无损编码在编码端，红蓝分量高频子带被丢弃而不参与编 码在解码端，先解码无损绿分量，再估计在红蓝位置的绿分量，其 高频子带可近似替代编码端丢弃的红蓝分量高频子带，最后反变换得到红 蓝分量的解码值 3、应用 LPA-ICI 方法构造了二阶非线性自适应滤波器，在对贝尔模板 像进行降噪插值的基础上，提出了迭代滤波法进行差信号后处理的方 法，通过迭代滤波可以有效地去除线性插值在估计红蓝像素时产生的边缘 伪彩噪声针对包含编码噪声的贝尔模板像，该方法有效地提高了重构 像质量 4、论文完成了 130 万像素的视频展台摄像头设计其中，采用 FPGA 实现了本文提出的贝尔模板无损编码器，可以替代基于 JPEG/JPEG2000 的 贝尔模板像无损编码方案，并集成自动聚焦和自动光圈控制模块，提高 了产品性能。

本文提出的贝尔模板像编解码算法及插值算法有利于简化数字成像 前端的硬件设计，降低功耗，在分布式无线像传输系统中具有很好的应 用前景 关键词：贝尔模板，像编码，像重构，差分预测，小波变换 太原理工大学博士研究生学位论文 III STUDY ON DEMOSAICING AND COMPRESSION ALGORITHM FOR BAYER IMAGES AND ITS APPLICATION ABSTRACT In recent years, Bayer image sensor has been widely applied in a variety of digital imaging devices, the relative image processing technologies and applications have become an important research field. Now, the algorithms of reconstructing full color image and encoding Bayer image are growing into focus of digital imaging technologies. In this paper, to reduce the complexity of front-end imaging devices, the temporal-frequency characteristic of Bayer images is analyzed first, then the correlation between demosaicing and compression algorithm is studied, at last, a novel coding method and corresponding demosaicing algorithm are proposed. Besides, part of the research results has been developed into products. The main innovation of this paper is as follows: 1. Based on international standards and frame work of JPEG-LS, an effective coder with causal interpolation module for Bayer image is subtly designed. Two replaceable predictors are improved to weaken correlation among pixels. Furthermore, to improve the coding gain, the quantized intervals of contexts are re-optimized. Consequently, the new coder can encode Bayer images pixel by pixel according to raster display sequence while maintains 太原理工大学博士研究生学位论文 IV balance between compression rate and complexity. The proposed coder can become an expanded part of JPEG-LS standard. 2. To obtain higher compression rate of Bayer pattern image, a visual lossless compression method that grounded on wavelet sub-band replacement is proposed. A differential prediction coder is employed to encode green component of Bayer images, which help preserving original intensity information of Bayer images. By using simple linear interpolation method, the green component at primary red or blue pixel positions is estimated, and low frequency sub-bands of these components are produced by utilizing DWT. Therefore, the green-red or the green-blue differential images of low frequency sub-bands are computed and encoded by using JPEG-LS in lossless or near-lossless mode without encoding their high frequency sub-bands. In coder, high frequency sub-bands of red and blue components are discarded, so they are not being coded. In decoder, lossless green component is firstly decoded, then green component at primary red or blue pixel positions is estimated, and its high frequency sub-bands are used to replace the information that discarded in the coder. Finally, the decoded pixels which are corresponding to the primary red or blue pixels are got by inverse transform. 3. A second-order nonlinear adaptive filter is constructed by LPA-ICI method. Based on noise reduction and interpolation treatment for Bayer pattern image, color differential component post-processing approach by iterative filter is proposed. The new design can effectively get rid of edge pseudo-color 太原理工大学博士研究生学位论文 V resulted from linear interpolation and preserve sharpness of reconstructed images, so as to improve the quality of reconstructed digital images. 4. Finally, a camera with 1.3M CCD used in visual presenter is developed. A key technology in this device is realization of Bayer images codec in a chip of FPGA with auto-focus and auto-iris modules, which can replace JPEG/JPEG2000 standard based coder of Bayer image and improve product performance. The fruit of this paper can simplify the hardware design and reduce power consumption of digital imaging devices, therefore, it has significant meaning in theoretic study and bright application future in distributed wireless image delivery system. KEY WORDS: bayer pattern, image code, demosaicing, differ。