Distributed Video Coding using Local Rank Transform

Abstract

In today's world, with advanced technology and easily available portable gadgets as enabler, video has become an important medium of communication. While, the standard H.264/AVC has produced extremely good results, it has not been suitable for this application domain as it is computationally intensive and unsuitable for low-power resources. In distributed video coding (DVC), an encoder requires less computation than the decoder, which usually runs at sites of higher computational resources. Our approach is to find out a novel approach in DVC, to reduce encoder complexity, using local rank transform (LRT). LRT relies on the relative ordering of local intensity values for application on visual correspondence problem. Use of LRT in DVC, to the best of our knowledge, has not been reported in literature before. First, we have developed techniques for image reconstruction using LRT and then design a DVC codec using it. We show analytically and by experimental results, that, in power-rate-distortion model, LRT encoder outperforms standard encoder (LDPCA in Stanford architecture) specially in low bit rate condition.

Low‐complexity feedback‐channel‐free distributed video coding using local rank transform

“…I 0 and I^ represent SI and reconstructed image, respectively. The option which produces lowest value of the cost function is used to update x. DLRTex [20] is the basic variant of DLRTexReg algorithm. DLRTex algorithm takes SI and original rank values of WZ frame as inputs.…”

“…To further reduce the bit rate, sampled LRT is used. We modify the decoding scheme reported in [8] to process the sampled LRT of an image. In the modified algorithm, the unknown rank pixel intensity values are recovered using SI and known rank pixel intensity values.…”

“…The proposed codec does not require any feedback channel, as we are coding all rank values independent on the response from the decoder. In the next section, we briefly discuss the base algorithm as reported in [8]. Subsequently we present the modifications proposed in this work for improving its performance.…”

“…In addition, mean pixel intensity of each block of WZ frame is sent to the decoder to assist ME, which improves the SI quality. DLRTexReg [19] is a complex algorithm compared with its basic variant DLRTex [20], as it needs to compute average value over each pixel neighbourhood in each iteration. In our approach, we use DLRTex with adaptive step size to decrease the decoder complexity, named adaptive DLRTex (ADLRTex).…”

“…Distributed video coding using LRT is first proposed in [8]. It reports less computational complexity in LRT encoder over conventional DVC methods like LDPCA and turbo codes.…”