Adaptive image sequence coding with neural network vector quantization

Summary form only given, as follows. A novel method for encoding an image sequence, termed adaptive neural net vector quantization (ANNVQ), has been devised. It is designed for use in encoding image sequences and based on Kohonen´s self-organizing feature maps, a neural-network-type clustering algorithm. It differs from it in that a modified form of adaptation resumes, after training the initial codebook, in order to respond to scene changes and motion. The main advantages are high image quality with modest bit rate and effective adaptation to motion and or scene changes, with the capability to quickly adjust the instantaneous bit rate in order to keep the image quality constant. This is a good match to packet-switched networks where variable bit rate and uniform image quality are highly desirable. Simulation experiments have been carried out with 4*4 blocks of pixels from an image sequence consisting of 20 frames of size 112*96 pixels each. With a codebook size of 512, ANNVQ results in high image quality upon image reconstruction, with peak signal-to-noise ratios of about 36-37 dB, at coding bit rates of about 0.50 b/pixel. This compares quite favorably with classical vector quantization of similar bit rates.<>