Image compression method

• US 5,313,534 A
• Filed: 08/25/1992
• Issued: 05/17/1994
• Est. Priority Date: 09/20/1991
• Status: Expired due to Fees

• Current Assignee: Thomson-CSF
• Original Assignee: Thomson-CSF
• Inventors: Burel, Gilles
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US07/936,199
• Time in Patent Office: 630 Days
• Field of Search: 382/56, 382/34, 358/433
• US Class Current: 382/238
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06T 9/002 using neural networks
• CPC Class Codes: H03M 7/3082 Vector coding for television signals, see H04N19/94
• CPC Class Codes: H04N 19/593 involving spatial prediction techniques
• CPC Class Codes: H04N 19/94 Vector quantisation

Abstract

A method for the compression of images comprises two steps. A first step of vector quantization compresses a source image into an image of indices that is coherent in relation to the source image. A second step again compresses this image of the indices according to a prediction method using a network of multilayer neurons. Application: compression of images for their storage and transmission.

Claims

1. A method for transmitting or storing a source image, said method using vector quantization, a prediction network, and an encoder to compress the source image, said method comprising the steps of

(a) dividing the source image into blocks of pixels,

(b) comparing each block with a plurality of vectors stored in a dictionary, each vector having an index,

(c) replacing each block of the source image with an index of the vector in the dictionary that is closest to the block of the image to obtain a compressed image, wherein blocks from the source image with similar content are replaced by indices located close to one another in the dictionary so that the compressed image is coherent with respect to the source image, and wherein i(x,y) is the index located at column x and row y of the compressed image,

(d) using a prediction network, generating a prediction i(x,y) for each index i(x,y) by receiving at an input of the prediction network a vector comprised of indices in the compressed image which are neighbors of i(x,y),

(e) transmitting a prediction error of each index, i(x,y)-i(x,y), to an encoder,

(f) at the encoder, encoding the prediction errors to generate encoded data, and

(g) transmitting the encoded data via transmission means or storing the encoded data in storage means.

2. The method of claim 1 wherein said prediction network comprises a plurality of multi-layer neurons.

3. The method of claim 1 wherein said prediction errors are encoded by said encoder using a Huffman code.