Vector quantizer

• US 4,558,350 A
• Filed: 06/13/1983
• Issued: 12/10/1985
• Est. Priority Date: 06/11/1982
• Status: Expired due to Term

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Inventors: Murakami, Tokumichi
• Primary Examiner(s): Murray, Richard
• Application Number:
  US06/503,473
• Time in Patent Office: 911 Days
• Field of Search: 358/21 R, 358/12, 358/36, 358/37, 358/133
• US Class Current: 375/240.12
• CPC Class Codes: G06T 9/008 Vector quantisation
• CPC Class Codes: H04N 11/042 Codec means
• CPC Class Codes: H04N 9/64 Circuits for processing colour signals H04N9/77 takes precedence; camera processing pipelines for processing colour signals H04N23/84

Abstract

A vector quantizer receives a plurality of input signals in block form and operates to determine an output vector having the least distortion with respect to an input vector by various techniques. A coded output or output vector codetable address is output from a coder of the device for receipt by a decoder which constructs a proper output vector signal.

Claims

1. A vector quantizer for processing color video signals, comprising;

an input vector register for receiving an input vector X composed of a block of K (K being more than one) samples including at least two color video signals from each of three channels;

an output vector code table memory for particitioning a signal space S^k of said input vector into subspaces having minimum distorted partition and which stores a set of the representative points of said subspaces produced as output vectors;

a mini-max arithmeticoperator that detects an output vector which, as a result of mini-max approximation, has been found to be the least distorted from the input vector from among the output vectors successively read from said codetable memory;

a coder for encoding the address of said least distorted output vector in said codetable and for delivering the coded address as an output signal;

a decoder for reading a corresponding output vector from a decoding output vector code memory using said coded output signal as an address signal; and

an output vector decomposing register that decomposes the output vector from said decoder into three channels of output signal sequence.

2. A d.c. corrected vector quantizer, comprising;

a codetable memory for partitioning a K-dimensional signal space including an input vector consisted of K input samples (K being an integer larger than

     1) minus the mean value of the block into subspaces having minimum distorted partitions with respect to the probability density of said input vectors and for storing the representative output vectors of said subspaces;

a mean value calculator for determining the mean of the block of K samples;

an adder for providing a d.c. corrected output vector by adding said block means to each component of the output vectors successively read from said codetable memory;

a maximum distortion detector for comparing said input vector with each of the d.c. corrected output vectors so as to detect the maximum absolute value of the difference as maximum distortion;

a minimum distorted vector detector for detecting an output vector having the least maximum distortion of the d.c. corrected output vectors successively compared with said input vector;

a coder for outputting a coded block mean and a coded address of said codetable corresponding to said least distorted output vector;

a decoder for receiving the output of said coder for decoding the address of said least distorted vector and the block mean;

a codetable memory for reading the least distorted output vector from the codetable with the aid of the address data delivered from said decoder; and

an adder for summing the block mean and the least distorted vector from the decoder and for delivering the sum as a d.c. corrected output vector.

3. A vector quantizer, comprising;

a normalized output vector codetable momory for partitioning a K-dimensional signal space including an input vector consisted of K (K being more than one) samples into subspaces having minimum distorted partitions with respect to the probability density function of an normalized input vector obtained by separating the d.c. component from said input vector and by normalizing its amplitude;

a mini-max arithmetic operator for normalizing said input vector, comparing it with each of the normalized output vectors successively read from said codetable memory, and detecting a normalized output vector that is the least distorted for said normalized input vectors;

a code circuit for encoding the d.c. component of said input vector and the normalizing constant thereof, as well as the index for said least distorted normalized output vector; and

a decoder for reconstructing an output vector which is the least distorted from said input vector using the d.c. component, the normalizing constant and the index for the least distorted normalized output vector, each of these three parameters being produced by decoding the output from said code circuit.

4. An interframe coding apparatus using a vector quantizer, comprising;

a frame memory for storing at least one frame of video signals at all times;

a subtractor for, upon receiving a block of the current K (K being more than one) input video samples, calculating interframe differential signals by reading from said memory a block of predictive signals which are at the same positions on a frame at least one frame before the current frame;

a limiter for suppressing small variations in and overload on said interframe differential signals;

a d.c. component separating and normalizing circuit for converting the suppressed interframe differential signals into input vectors by separating the mean value for the block of the signals therefrom and by normalizing said signals using a component of the standard deviation from said mean value;

a motion detecting circuit for comparing said mean and standard deviation components for a given block with a predetermined threshold value and which, upon finding that the two parameters are less than said threshold value, interprets said block as an insignificant block and performs an operation that cancels all of the interframe differential signals in that block;

a coding unit for vector quantization for encoding the input vectors in a significant block, the block mean or standard deviation component thereof being greater than the threshold value, into a code for distinguishing an output vector which has the minimum distortion, defined by the smallest distance in a K-dimensional signal space R^k, from the input vector among a predetermined number of representative points of a preselected optimum arrangement of output vectors within the K-dimensional signal space R^k ;

a decoding unit for vector quantization for selecting a corresponding output vector from said distinguishing code, multiplying the same by the block standard deviation component and adding the block mean to the product, for reproducing the set of interframe differential signals, and for further nullifying the interframe differential signals in a insignificant block;

an adder which, after said vector quantization, adds the predictive signals to the reproduced interframe differential signals to reproduce video signals and for then writing said video signals into said frame memory.

5. An interframe coding apparatus according to claim 4, and further including a transmission data buffer for performing variable word length coding on said code for distinguishing between significant and insignificant blocks, the code for distinguishing the least distorted output vector, and the mean value and standard deviation of said significant block and for controlling the threshold value, for keeping the amount of information at a constant level.

6. An interframe coding apparatus according to claim 4, wherein the component of standard deviation from the block mean for normalizing or detecting significant motions in said interframe differential signals comprises the difference between the maximum and minimum values for the block or the total sum of the absolute values of the differences from the block mean.

7. An interframe coding apparatus according to claim 4, wherein said coding unit for vector quantization includes a section for calculating the distortion between input and output vectors defined by the distance between the two vectors in a K-dimensional signal space in maximum terms of the absolute value of the greatest difference between the elements of the input and output vectors, or in terms of the toal sum of the absolute values of the differences between the elements of the input and output vectors.