Reference Frame Compression Using Embedded Reconstruction Patterns for H.264/AVC Decoder

This paper presents a novel method of reference frame compression for use in H.264/AVC video codecs. The method comprises a predictive pattern decision for encoding of original 2times2 pixel blocks, selective step quantization and fast reconstruction of embedded patterns during the decoding. The proposed approach has a number of advantages over existing methods, such as random access to compressed data, a constant compression rate of 50% and low cost hardware-friendly implementation requiring less than 7 k gates. Experimental results indicate an acceptable quality loss of 0.47 dB on average for video resolutions from CIF to 720p and only 5-15% increase in total computational complexity for the software implementation.