压缩感知以部分随机变换代替全变换,仍可确保图像的高精度复原,可用于铁路系统中无线监控终端实现低复杂图像编码.传统图像压缩感知以相同的测量率实施分块测量,而由于分块稀疏度不同,常导致重建图像具有块效应,无法确保良好的率失真性能.为了解决该问题,本文提出利用图像纹理特征引导图像压缩感知,在感知端实施自适应测量.利用像素八连通区域内的最大梯度度量各像素的纹理变化程度,生成纹理特征图,利用纹理特征图计算各块纹理对比度,并以此为依据自适应设定各块的测量率,以块纹理对比度加权图像重建模型的目的函数,集中优化纹理细节的区域.实验结果表明,与由块方差、边缘特征主导的自适应测量方法相比,本文所提算法可确保较好的重建图像主观视觉质量,且率失真性能优于传统压缩感知重建算法.%Compressive sensing(CS) can reconstruct the image with a high accuracy by exploiting the partial random transformation instead of full transformation,which can be applied in the wireless monitor terminal of the rail system to realize the low-complexity encoding.The traditional image CS conducts block-wise measuring with the same measurement rate,causing some blocking artifacts of reconstructed image due to the different block sparsity,which results in a low rate-distortion(RD) performance.To solve the above problem,this paper proposed adaptively measuring each block depending on the texture features of image in the sensor.Firstly,the maximum gradient in the 8-connected region of each pixel was used to represent the corresponding texture variation,and these gradient values constructed the texture-feature map.Then,the texture contrast of each block was computed by using the texture-feature map,to adaptively set the measurement rate of each block according to the distribution of texture contrast.Finally,the objective function of image reconstruction model was weighted by the texture contrast to focus on the optimization of rich texture region.Experimental results showed that the proposed algorithm can guarantee the better subjective visual quality of the reconstructed image when compared with the adaptive measurement methods oriented with block variance and edge features,and the RD performance of the proposed algorithm outperformed that of the traditional CS reconstruction algorithm.
展开▼
