A novel approach for near-lossless compression of Color Filtering Array (CFA) data in wireless endoscopy capsule is proposed in this paper. The compression method is based on pre-processing and vector quantization. Fi...A novel approach for near-lossless compression of Color Filtering Array (CFA) data in wireless endoscopy capsule is proposed in this paper. The compression method is based on pre-processing and vector quantization. First, the CFA raw data are low pass filtered and rearranged during pre-processing. Then, pairs of pixels are vector quantized into macros of 9 bits by applying block par-tition and index mapping in succession. These macros are entropy compressed by Joint Photographic Experts Group-Lossless Standard (JPEG-LS) finally. The complex step of codeword searching in Vector Quantization (VQ) is avoided by a predefined partition rule, which is suitable for hardware imple-mentation. By control of the pre-processor and VQ scheme, either high quality compression under un- filtered case or high ratio compression under filtered case can be realized, with the average Peak Sig-nal-to-Noise Ratio (PSNR) more than 43dB and 37dB respectively. Compared with the state-of-the-art method and the previously proposed method, our compression approach outperforms in compression performance as well as in flexibility.展开更多
在混合激励线性预测(mixed excitation linear prediction,MELP)模型的基础上,以超帧为单位,采用多帧联合编码技术,分模式对子帧的语音特征参数进行联合量化,实现了一种码率为600 bit/s的声码器。为了进一步减小量化误差,设计出了一种...在混合激励线性预测(mixed excitation linear prediction,MELP)模型的基础上,以超帧为单位,采用多帧联合编码技术,分模式对子帧的语音特征参数进行联合量化,实现了一种码率为600 bit/s的声码器。为了进一步减小量化误差,设计出了一种基于高斯混合模型的预测分类分裂矢量量化器(predictive switched split vector quantization based on Gauss mixture model,GMM-PSSVQ),该量化器对超帧中某些子帧的线谱频率进行量化,并利用帧间预测和线性插值等方法提高编码效率。采用谱失真对设计的矢量量化器进行性能评估,并分别与多级矢量量化和预测分裂矢量量化算法进行性能比较;通过客观感知语音质量评估和主观判断韵字测试对实现的声码器进行性能测试。测试结果表明,设计的矢量量化器平均谱失真最低,实现的声码器合成语音具有较高的清晰度和可懂度。展开更多
基金the National Natural Science Foundation of China (No. 60506007).
文摘A novel approach for near-lossless compression of Color Filtering Array (CFA) data in wireless endoscopy capsule is proposed in this paper. The compression method is based on pre-processing and vector quantization. First, the CFA raw data are low pass filtered and rearranged during pre-processing. Then, pairs of pixels are vector quantized into macros of 9 bits by applying block par-tition and index mapping in succession. These macros are entropy compressed by Joint Photographic Experts Group-Lossless Standard (JPEG-LS) finally. The complex step of codeword searching in Vector Quantization (VQ) is avoided by a predefined partition rule, which is suitable for hardware imple-mentation. By control of the pre-processor and VQ scheme, either high quality compression under un- filtered case or high ratio compression under filtered case can be realized, with the average Peak Sig-nal-to-Noise Ratio (PSNR) more than 43dB and 37dB respectively. Compared with the state-of-the-art method and the previously proposed method, our compression approach outperforms in compression performance as well as in flexibility.
文摘在混合激励线性预测(mixed excitation linear prediction,MELP)模型的基础上,以超帧为单位,采用多帧联合编码技术,分模式对子帧的语音特征参数进行联合量化,实现了一种码率为600 bit/s的声码器。为了进一步减小量化误差,设计出了一种基于高斯混合模型的预测分类分裂矢量量化器(predictive switched split vector quantization based on Gauss mixture model,GMM-PSSVQ),该量化器对超帧中某些子帧的线谱频率进行量化,并利用帧间预测和线性插值等方法提高编码效率。采用谱失真对设计的矢量量化器进行性能评估,并分别与多级矢量量化和预测分裂矢量量化算法进行性能比较;通过客观感知语音质量评估和主观判断韵字测试对实现的声码器进行性能测试。测试结果表明,设计的矢量量化器平均谱失真最低,实现的声码器合成语音具有较高的清晰度和可懂度。
