Image fidelity estimation from received embedded bitstream

Abstract

Despite the fact that mean squared error (MSE) is the most widely used image/video fidelity criterion, it requires the availability of original image and large computational resources for its computation. The non-availability of original images limits the use of MSE as a quality metric in many practical scenarios. This paper proposes to use MSE reduction (MSER) with reference to null image for estimating the fidelity of images decoded from embedded bitstreams without the original image. MSER is generally used for rate allocation, optimal error protection, etc. However, MSER gives information about the improvement in fidelity of decoded image, rather than its absolute value. It is proposed to estimate MSE from $${\textit{MSER}}_f$$ (MSE reduction in lossless bitstream) and $${\textit{MSER}}_d$$ (MSE reduction at desired bitrate/fidelity). The $${\textit{MSER}}_d$$ can be calculated at decoder side while decoding the embedded bitstream at desired rate. $${\textit{MSER}}_f$$ is fixed and independent of bitrate, calculated at the encoder side and is embedded in the bitstream as side information. It has been suggested that knowing the values of $${\textit{MSER}}_f$$ and $${\textit{MSER}}_d$$, the MSE of decoded image is simply their difference. Simulation results demonstrate that the proposed method is fairly simple and highly accurate.