VPIC-based image compression technique for sensor network applications

This paper examines the implementation considerations of visual pattern image coding (VPIC) algorithm for image compression on the focal plane. Several basic visual patterns that exploit the psychovisual redundancy are adopted in the VPIC coding. These patterns are essentially the same edge patterns developed in classified vector quantization (CVQ). The isometry operation is introduced to help eliminate the gradient angle and edge polarity calculation by exhaustive pattern search. By implementing the time-to-first spike (TFS) digital pixel together with the VPIC coding, the feature that TFS pixels are naturally sorted could be exploited to facilitate and simplify the block mean and gradient magnitude calculation. The reconstructed image quality in terms of Peak Signal-to-Noise Ratio (PSNR) for lena image at the 0.875 bpp could be around 29 dB. This technique is very suitable for Wireless Sensor Network (WSN) applications where image quality is not the primary concern.