Error Control and Regression for Block Compressed Sensing of Color Images

Transmission of multimedia contents with block compressed sensing (BCS) would be of great value for practical applications. Compressed data are vulnerable to channel errors during transmission, and effective means to protect them would be helpful to the decoding procedures. We choose multiple description coding (MDC) for error control in advance, and apply the least absolute shrinkage and selection operator (LASSO) for training the received data to alleviate the effect caused by channel errors. We also consider the inherent characteristics between color planes of original image to enhance the reconstruction. Simulations have depicted the advantages with these error control schemes for BCS.

I. Introduction

Block compressed sensing (BCS) [1] [2], which follows the concept of block transform coding like discrete cosine transform (DCT), is an evolving topic in compressed sensing (CS) researches [3]. The input image is divided into blocks with the size of , and each block is applied with compressed sensing independently. Conventional researches mainly lie into looking for compression performances with the rate-distortion analysis, and we focus on error-prone transmission for BCS. During transmission, parts of the BCS coefficients may be affected by channel error or data loss, which may have negative effects to the data reconstruction at the decoder. Hence, error control schemes would be much required to be applied to the BCS coefficients at the encoder. With these observations, we employ multiple description coding (MDC), an error control scheme, at the encoder to mitigate the effects cause by the channels, and choose the least absolute shrinkage and selection operator (LASSO) to train the regression of received BCS coefficients at the decoder. We expect to see the results with error control and regression that would surpass with those without protection.