In this paper a novel coding method based on fuzzy vector quantization for noised image with Gaussian white-noise pollution is presented. By restraining the high frequency subbands of wavelet image the noise is signif...In this paper a novel coding method based on fuzzy vector quantization for noised image with Gaussian white-noise pollution is presented. By restraining the high frequency subbands of wavelet image the noise is significantly removed and coded with fuzzy vector quantization. The experimental result shows that the method can not only achieve high compression ratio but also remove noise dramatically.展开更多
A fast encoding algorithm based on the mean square error （MSE） distortion for vector quantization is introduced. The vector, which is effectively constructed with wavelet transform （WT） coefficients of images, can...A fast encoding algorithm based on the mean square error （MSE） distortion for vector quantization is introduced. The vector, which is effectively constructed with wavelet transform （WT） coefficients of images, can simplify the realization of the non-linear interpolated vector quantization （NLIVQ） technique and make the partial distance search （PDS） algorithm more efficient. Utilizing the relationship of vector L2-norm and its Euclidean distance, some conditions of eliminating unnecessary codewords are obtained. Further, using inequality constructed by the subvector L2-norm, more unnecessary codewords are eliminated. During the search process for code, mostly unlikely codewords can be rejected by the proposed algorithm combined with the non-linear interpolated vector quantization technique and the partial distance search technique. The experimental results show that the reduction of computation is outstanding in the encoding time and complexity against the full search method.展开更多
With the advances of display technology, three-dimensional(3-D) imaging systems are becoming increasingly popular. One way of stimulating 3-D perception is to use stereo pairs, a pair of images of the same scene acqui...With the advances of display technology, three-dimensional(3-D) imaging systems are becoming increasingly popular. One way of stimulating 3-D perception is to use stereo pairs, a pair of images of the same scene acquired from different perspectives. Since there is an inherent redundancy between the images of a stereo pairs, data compression algorithms should be employed to represent stereo pairs efficiently. The proposed techniques generally use block-based disparity compensation. In order to get the higher compression ratio, this paper employs the wavelet-based mixed-resolution coding technique to incorporate with SPT-based disparity-compensation to compress the stereo image data. The mixed-resolution coding is a perceptually justified technique that is achieved by presenting one eye with a low-resolution image and the other with a high-resolution image. Psychophysical experiments show that the stereo image pairs with one high-resolution image and one low-resolution image provide almost the same stereo depth to that of a stereo image with two high-resolution images. By combining the mixed-resolution coding and SPT-based disparity-compensation techniques, one reference (left) high-resolution image can be compressed by a hierarchical wavelet transform followed by vector quantization and Huffman encoder. After two level wavelet decompositions, for the low-resolution right image and low-resolution left image, subspace projection technique using the fixed block size disparity compensation estimation is used. At the decoder, the low-resolution right subimage is estimated using the disparity from the low-resolution left subimage. A full-size reconstruction is obtained by upsampling a factor of 4 and reconstructing with the synthesis low pass filter. Finally, experimental results are presented, which show that our scheme achieves a PSNR gain (about 0.92dB) as compared to the current block-based disparity compensation coding techniques.展开更多
[目的/意义]为实现山楂水分含量的快速无损检测,本研究探索了一种基于高光谱成像技术结合机器学习算法的检测方法。[方法]首先,收集458个来自不同产区不同品种的新鲜山楂样品,分别采集每个样品在可见-近红外波段(Visible to Near Infrar...[目的/意义]为实现山楂水分含量的快速无损检测,本研究探索了一种基于高光谱成像技术结合机器学习算法的检测方法。[方法]首先,收集458个来自不同产区不同品种的新鲜山楂样品,分别采集每个样品在可见-近红外波段(Visible to Near Infrared,VNIR)和短波红外(Short-Wave Infrared,SWIR)波段的高光谱数据,利用阈值分割算法确定每个山楂的感兴趣区域(Region of Interest,ROI),提取果实ROI的平均反射光谱作为原始数据。随后,采用卷积平滑、乘法散射校正、标准正态变换、一阶导数和二阶导数五种预处理方法,对原始光谱数据进行优化。在此基础上,结合偏最小二乘回归、支持向量回归(Support Vector Regression,SVR)、随机森林与多层感知机等机器学习方法,系统评估不同摆放方式(果柄朝侧面、朝上、朝下及三者融合)和光谱范围(VNIR、SWIR、VNIR+SWIR)对模型预测性能的影响。最后,采用连续投影算法、竞争自适应重加权采样算法、变量迭代空间收缩方法,以及离散小波变换-逐步回归(Discrete Wavelet Transform-Stepwise Regression,DWT-SR)四种方法对全波段数据进行降维处理,进一步减少数据冗余,提高模型效率。[结果和讨论]果柄朝下的摆放方式、SWIR波段范围(940~2500 nm)及一阶导数预处理组合下,SVR模型表现最优,测试集的绝对系数(Coefficient of Determination,R^(2)_(p))为0.8605、平均绝对误差(Mean Absolute Error,MAE p)为0.7111、均方根误差(Root Mean Square Error,RMSE p)为0.9142、相对分析误差(Ratio of Performance to Deviation,RPD)为2.6776。在性能最优分析条件下,DWT-SR方法基于小波基函数“db6”在分解层级为1时,提取出17个关键特征波段,所建模型在降低数据维度的同时可以保持高水平预测性能(R^(2)_(p)=0.8571、MAE_(p)=0.6692、RMSE p=0.9252、RPD=2.6457)。[结论]本研究证明了高光谱成像结合机器学习方法在山楂水分无损检测中的可行性,为果品水分在线监测及智能分选提供了理论依据与技术支撑。展开更多
This paper presents a new method for image coding and compressing-ADCTVQ(Adptive Discrete Cosine Transform Vector Quantization). In this method, DCT conforms to visual properties and has an encoding ability which is i...This paper presents a new method for image coding and compressing-ADCTVQ(Adptive Discrete Cosine Transform Vector Quantization). In this method, DCT conforms to visual properties and has an encoding ability which is inferior only to the best transform KLT. Its vector quantization can maintain the minimum quantization distortions and greatly increase the compression ratio. In order to improve compression efficiency, an adaptive strategy of selecting reserved region patterns is applied to preserving the high energy at the same compression ratio. The experiment results show that they are satisfactory at the compression ration ratio if greater than 20.展开更多
随着人工智能的发展,深度神经网络成为多种模式识别任务中必不可少的工具,由于深度卷积神经网络(CNN)参数量巨大、计算复杂度高,将它部署到计算资源和存储空间受限的边缘计算设备上成为一项挑战。因此,深度网络压缩成为近年来的研究热...随着人工智能的发展,深度神经网络成为多种模式识别任务中必不可少的工具,由于深度卷积神经网络(CNN)参数量巨大、计算复杂度高,将它部署到计算资源和存储空间受限的边缘计算设备上成为一项挑战。因此,深度网络压缩成为近年来的研究热点。低秩分解与向量量化是深度网络压缩中重要的两个研究分支,其核心思想都是通过找到原网络结构的一种紧凑型表达,从而降低网络参数的冗余程度。通过建立联合压缩框架,提出一种基于低秩分解和向量量化的深度网络压缩方法——可量化的张量分解(QTD)。该方法能够在网络低秩结构的基础上实现进一步的量化,从而得到更大的压缩比。在CIFAR-10数据集上对经典ResNet和该方法进行验证的实验结果表明,QTD能够在准确率仅损失1.71个百分点的情况下,将网络参数量压缩至原来的1%。而在大型数据集ImageNet上把所提方法与基于量化的方法PQF(Permute,Quantize,and Fine-tune)、基于低秩分解的方法TDNR(Tucker Decomposition with Nonlinear Response)和基于剪枝的方法CLIP-Q(Compression Learning by In-parallel Pruning-Quantization)进行比较与分析的实验结果表明,QTD能够在相同压缩范围下实现更好的分类准确率。展开更多
文摘In this paper a novel coding method based on fuzzy vector quantization for noised image with Gaussian white-noise pollution is presented. By restraining the high frequency subbands of wavelet image the noise is significantly removed and coded with fuzzy vector quantization. The experimental result shows that the method can not only achieve high compression ratio but also remove noise dramatically.
基金the National Natural Science Foundation of China (60602057)the NaturalScience Foundation of Chongqing Science and Technology Commission (2006BB2373).
文摘A fast encoding algorithm based on the mean square error （MSE） distortion for vector quantization is introduced. The vector, which is effectively constructed with wavelet transform （WT） coefficients of images, can simplify the realization of the non-linear interpolated vector quantization （NLIVQ） technique and make the partial distance search （PDS） algorithm more efficient. Utilizing the relationship of vector L2-norm and its Euclidean distance, some conditions of eliminating unnecessary codewords are obtained. Further, using inequality constructed by the subvector L2-norm, more unnecessary codewords are eliminated. During the search process for code, mostly unlikely codewords can be rejected by the proposed algorithm combined with the non-linear interpolated vector quantization technique and the partial distance search technique. The experimental results show that the reduction of computation is outstanding in the encoding time and complexity against the full search method.
基金This project was supported by the National Natural Science Foundation (No. 69972027).
文摘With the advances of display technology, three-dimensional(3-D) imaging systems are becoming increasingly popular. One way of stimulating 3-D perception is to use stereo pairs, a pair of images of the same scene acquired from different perspectives. Since there is an inherent redundancy between the images of a stereo pairs, data compression algorithms should be employed to represent stereo pairs efficiently. The proposed techniques generally use block-based disparity compensation. In order to get the higher compression ratio, this paper employs the wavelet-based mixed-resolution coding technique to incorporate with SPT-based disparity-compensation to compress the stereo image data. The mixed-resolution coding is a perceptually justified technique that is achieved by presenting one eye with a low-resolution image and the other with a high-resolution image. Psychophysical experiments show that the stereo image pairs with one high-resolution image and one low-resolution image provide almost the same stereo depth to that of a stereo image with two high-resolution images. By combining the mixed-resolution coding and SPT-based disparity-compensation techniques, one reference (left) high-resolution image can be compressed by a hierarchical wavelet transform followed by vector quantization and Huffman encoder. After two level wavelet decompositions, for the low-resolution right image and low-resolution left image, subspace projection technique using the fixed block size disparity compensation estimation is used. At the decoder, the low-resolution right subimage is estimated using the disparity from the low-resolution left subimage. A full-size reconstruction is obtained by upsampling a factor of 4 and reconstructing with the synthesis low pass filter. Finally, experimental results are presented, which show that our scheme achieves a PSNR gain (about 0.92dB) as compared to the current block-based disparity compensation coding techniques.
文摘[目的/意义]为实现山楂水分含量的快速无损检测,本研究探索了一种基于高光谱成像技术结合机器学习算法的检测方法。[方法]首先,收集458个来自不同产区不同品种的新鲜山楂样品,分别采集每个样品在可见-近红外波段(Visible to Near Infrared,VNIR)和短波红外(Short-Wave Infrared,SWIR)波段的高光谱数据,利用阈值分割算法确定每个山楂的感兴趣区域(Region of Interest,ROI),提取果实ROI的平均反射光谱作为原始数据。随后,采用卷积平滑、乘法散射校正、标准正态变换、一阶导数和二阶导数五种预处理方法,对原始光谱数据进行优化。在此基础上,结合偏最小二乘回归、支持向量回归(Support Vector Regression,SVR)、随机森林与多层感知机等机器学习方法,系统评估不同摆放方式(果柄朝侧面、朝上、朝下及三者融合)和光谱范围(VNIR、SWIR、VNIR+SWIR)对模型预测性能的影响。最后,采用连续投影算法、竞争自适应重加权采样算法、变量迭代空间收缩方法,以及离散小波变换-逐步回归(Discrete Wavelet Transform-Stepwise Regression,DWT-SR)四种方法对全波段数据进行降维处理,进一步减少数据冗余,提高模型效率。[结果和讨论]果柄朝下的摆放方式、SWIR波段范围(940~2500 nm)及一阶导数预处理组合下,SVR模型表现最优,测试集的绝对系数(Coefficient of Determination,R^(2)_(p))为0.8605、平均绝对误差(Mean Absolute Error,MAE p)为0.7111、均方根误差(Root Mean Square Error,RMSE p)为0.9142、相对分析误差(Ratio of Performance to Deviation,RPD)为2.6776。在性能最优分析条件下,DWT-SR方法基于小波基函数“db6”在分解层级为1时,提取出17个关键特征波段,所建模型在降低数据维度的同时可以保持高水平预测性能(R^(2)_(p)=0.8571、MAE_(p)=0.6692、RMSE p=0.9252、RPD=2.6457)。[结论]本研究证明了高光谱成像结合机器学习方法在山楂水分无损检测中的可行性,为果品水分在线监测及智能分选提供了理论依据与技术支撑。
文摘This paper presents a new method for image coding and compressing-ADCTVQ(Adptive Discrete Cosine Transform Vector Quantization). In this method, DCT conforms to visual properties and has an encoding ability which is inferior only to the best transform KLT. Its vector quantization can maintain the minimum quantization distortions and greatly increase the compression ratio. In order to improve compression efficiency, an adaptive strategy of selecting reserved region patterns is applied to preserving the high energy at the same compression ratio. The experiment results show that they are satisfactory at the compression ration ratio if greater than 20.
文摘随着人工智能的发展,深度神经网络成为多种模式识别任务中必不可少的工具,由于深度卷积神经网络(CNN)参数量巨大、计算复杂度高,将它部署到计算资源和存储空间受限的边缘计算设备上成为一项挑战。因此,深度网络压缩成为近年来的研究热点。低秩分解与向量量化是深度网络压缩中重要的两个研究分支,其核心思想都是通过找到原网络结构的一种紧凑型表达,从而降低网络参数的冗余程度。通过建立联合压缩框架,提出一种基于低秩分解和向量量化的深度网络压缩方法——可量化的张量分解(QTD)。该方法能够在网络低秩结构的基础上实现进一步的量化,从而得到更大的压缩比。在CIFAR-10数据集上对经典ResNet和该方法进行验证的实验结果表明,QTD能够在准确率仅损失1.71个百分点的情况下,将网络参数量压缩至原来的1%。而在大型数据集ImageNet上把所提方法与基于量化的方法PQF(Permute,Quantize,and Fine-tune)、基于低秩分解的方法TDNR(Tucker Decomposition with Nonlinear Response)和基于剪枝的方法CLIP-Q(Compression Learning by In-parallel Pruning-Quantization)进行比较与分析的实验结果表明,QTD能够在相同压缩范围下实现更好的分类准确率。
