Method of codebook generation for an amplitude-adaptive vector quantization system

#The present invention relates to the field of vector quantization of transmitted imagery and, more particularly, is directed to a method for automatically generating a codebook for use in the transmitter and receiver components of an amplitude-adaptive vector quantization system based upon a specified signal-to-noise ratio goal. This signal-to-noise ratio goal serves as a convenient single control of the trade-off between received image fidelity and data compression. The larger the ratio, the greater the fidelity, but the less compression.

#Vector quantization as used in the transmission of imagery consists of dividing an image to be transmitted into rectangular blocks, or vectors, of pixels, referred to as the "starting image vectors." A vector quantization system generally comprises a transmitter unit and a receiver unit. The transmitter compares each starting image vector with a large library of typical vectors, called a "codebook," and selects the codebook vector that best approximates the starting image vector. The transmitter then encodes and transmits a codebook index corresponding to the selected codebook vector. The receiver, equipped with a copy of the codebook, decodes the index, retrieves the selected library vector and inserts it into the output image.

#Differential vector quantization (DVQ) involves first obtaining a mean value of the starting image vector (referred to as the vector mean or image block mean) by averaging the pixel elements that make up the starting image vector, and rounding the average to the nearest integer. A difference vector is then obtained by subtracting the vector mean from each pixel in the string of pixels that make up the starting image vector. The vector mean is modulated, such as by pulse code modulation (PCM) or by difference pulse code modulation (DPCM), and is transmitted to a companion receiver unit, while the difference vector is used to obtain and transmit a corresponding vector index from the codebook. Normalized differential vector quantization (NDVQ) is DVQ with the codebook vectors normalized such that their magnitudes always equal 1. This method of vector quantization provides higher vector data compression than does DVQ, but requires the separate transmission of the difference vector magnitude.

#The invention will become more readily apparent from the following exemplary description taken in conjunction with the accompanying drawings, in which:

#FIG. 1 is a block diagram of an amplitude-adaptive NDVQ transmitter, as known in the art;

#FIG. 2 is a block diagram of an amplitude-adaptive NDVQ receiver, also as known in the art, for use with the transmitter of FIG. 1.

#The basic units of an amplitude-adaptive NDVQ system are shown for the transmitter in FIG. 1 and for the receiver in FIG. 2. The transmitter unit of FIG. 1 comprises a processor 12 that receives a starting image vector or image block on line 10. One line output 13 from processor 12 is connected to a modulator 14 and provides an output m corresponding to the vector (image block) mean. Modulator 14 outputs a modulated signal m' which corresponds to the transmitted value of m and is transmitted to the receiver of FIG. 2. A second "difference vector" output at line 15 from processor 12 is connected to a codebook accessor 16 which, in turn, is connected to receive data from a codebook 18 that comprises a library of codebook vectors addressed by vector indices. Codebook accessor 16 is connected by an output line 17 to a further modulator 20. The input to modulator 20 is a signal p corresponding to the dot product of the difference and codebook vectors, and modulator 20 produces a modulated signal p' that corresponds to the transmitted value of p and is also transmitted to the receiver of FIG. 2. A second output line 19 from codebook accessor 16 is connected to a vector index corrector 22. The input to the vector index corrector 22 is a signal i corresponding to the codebook vector index, and the vector index corrector outputs a corrected codebook vector index signal i' for transmission to the receiver. The signal p' from modulator 20 is also provided over an output line 24 to a second input of vector index corrector 22.

#The receiver unit shown in FIG. 2 comprises a table lookup 32 that receives transmitted signals p' and i', and is also connected to receive data from a codebook 31. The codebook 31 in the receiver is a duplicate of the codebook 18 in the transmitter. The output of table lookup 32 is connected over line 33 to multiplier 34 which also receives signal p' and which outputs a signal over line 35 to adder 36. Adder 36 receives both this signal and the transmitted signal m', and provides at an output line 38 an output image vector.

#The transmission of signals between the transmitter and receiver is performed using known techniques, such as by direct wire connection or through radio frequency transmission.