Analog hardware implementation of a vector quantizer for focal-plane image compression

Abstract

We propose the use of a low-complexity vector quantizer for coding data vectors at the focal plane of CMOS image sensors, directly from analog pixels samples prior to A/D conversion. To be suitable for focal-plane image compression applications, the encoder must have a very low transistor count. Without considering the implementation errors in DPCM preprocessing of the image data, a test image is compressed by the proposed vector quantizer with peak signal-to-noise ratio around 26 dB at 0.73 bpp. The vector quantizer requires 145 transistors for each block of 4x4 pixels, and Monte Carlo simulation results in Cadence/Spectre led to the same image compression results that had been achieved in theoretical predictions.