Image recognition method and apparatus

• US 5,805,727 A
• Filed: 12/05/1995
• Issued: 09/08/1998
• Est. Priority Date: 12/05/1994
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Nakano, Hiroki
• Primary Examiner(s): Couso, Jose L.
• Assistant Examiner(s): NGUYEN, HA T
• Application Number:
  US08/567,221
• Time in Patent Office: 1,008 Days
• Field of Search: 382/192, 382/209, 382/217, 382/218, 382/248, 382/250, 382/299, 382/253, 382/278, 382/235, 382/240, 382/181, 382/195
• US Class Current: 382/195
• CPC Class Codes: G06T 9/005 Statistical coding, e.g. Huffman, run length coding
• CPC Class Codes: G06V 30/2504 Coarse or fine approaches, e.g. resolution of ambiguities or multiscale approaches

Abstract

A non-compressed template image is compared with compressed object images stored in a memory to find an object image corresponding to the template image. Each object image is DCT transformed and compressed according to JPEG algorithm such that the compressed image data includes DCT coefficients representing DC components which indicate average intensity or brightness of the respective block of the object image. The template image is processed to obtain average intensity for each of blocks corresponding in size to the blocks of the object image. Image recognition is performed by rough matching and fine matching. In the rough matching, the average intensity of the template image and the average intensity of the object image represented by the DC components are used to examine the degree of matching. When a maximum rough matching value is found, the object image is restored or decompressed and the fine matching is done with use of pixel intensity. The rough matching is carried out using DC components without restoration of the compressed image which involves an enormous amount of computations, and only the compressed image with the maximum rough matching value is restored. This reduces computation load and increases process speed.

Claims

1. An image recognition method for detecting matching between a non-compressed image and a compressed image, said compressed image being stored in memory as image data which are compressed using orthogonal transform into a two-dimensional spatial frequency region, said compressed image data including, for each of blocks each comprising a predetermined number of pixels, a DC component corresponding to the average intensity of the respective block, said image recognition method comprising the steps of:

dividing said non-compressed image into blocks corresponding in size to said blocks in said compressed image;

computing the average intensity for each of the blocks of said non-compressed image;

sequentially extracting DC components for a plurality of partial areas from said compressed image data, said partial areas corresponding in size to said non-compressed image;

computing a rough matching value indicative of the degree of matching between said non-compressed image and said compressed image based on said DC components of said partial areas and the average intensity of said blocks in said non-compressed image;

determining a candidate area based on said rough matching value and restoring compressed image data corresponding to said candidate area to obtain candidate image data;

computing a fine matching value indicative of the degree of matching between said candidate image and said non-compressed image based on the image data for said candidate image and the image data for said non-compressed image; and

determining based on said fine matching value whether or not said non-compressed image matches said compressed image.

2. The image recognition method according to claim 1, wherein said rough matching value is computed based on the correlation between each of said partial areas and said non-compressed image with respect to the average intensity by the block.

3. The image recognition method according to claim 1, wherein said fine matching value is computed based on the correlation between said candidate image and said non-compressed image with respect to the intensity by the pixel.

4. The image recognition method according to claim 1, wherein said intensity is the intensity of a predetermined color.

5. The image recognition method according to claim 1, further including the step of predetermining a reference value for each of said rough and fine matching, and the step of determining that the images do not match if at least one of said rough and fine matching values is smaller than the corresponding reference value.

6. The image recognition method according to claim 1, wherein the intensity of the compressed image is normalized.

7. The image recognition method according to claim 1, wherein said candidate area is restored correspondingly to a preset spatial frequency region.

8. An image recognition apparatus which detects matching between a non-compressed image and a compressed image, said compressed image being stored in memory as image data which are compressed using orthogonal transform into a two-dimensional spatial frequency region, said compressed image data including, for each of blocks each comprising a predetermined number of pixels, a DC component corresponding to the average intensity of the respective block, said image recognition apparatus comprising:

means for dividing said non-compressed image into blocks corresponding in size to said blocks in said compressed image,

means for computing the average intensity for each of the blocks of said non-compressed image,

means for sequentially extracting DC components for a plurality of partial areas from said compressed image data, said partial areas corresponding in size to said non-compressed image;

means for computing a rough matching value indicative of the degree of matching between said non-compressed image and said compressed image based on said DC components of said partial areas and the average intensity of said blocks in said non-compressed image;

means for determining a candidate area based on said rough matching value and for restoring compressed image data corresponding to said candidate area to obtain candidate image data;

means for computing a fine matching value indicative of the degree of matching between said non-compressed image and said candidate image based on the image data for said candidate image and the image data for said non-compressed image; and

means for determining based on said fine matching value whether or not said non-compressed image matches said compressed image.

9. An image recognition method for detecting matching between a non-compressed image and a compressed image, said compressed image being stored in memory as image data which are compressed using orthogonal transform into a two-dimensional spatial frequency region, said compressed image data including, for each of blocks each comprising a predetermined number of pixels in the original uncompressed image, a DC component corresponding to the average intensity of the respective block, said image recognition method comprising the steps of:

dividing said non-compressed image into one or more blocks corresponding in size to said blocks in said compressed image;

computing the average intensity for each of the blocks of said non-compressed image;

extracting DC components from said compressed image;

performing rough matching between said non-compressed image and said compressed image based on the DC components extracted from said compressed image and the average intensity computed for each of the blocks of said non-compressed image;

restoring the compressed image when the degree of matching in the rough matching

exceeds a predetermined threshold;

performing a fine matching between said non-compressed image and said restored image based on the image data for said restored image and the image data for said non-compressed image; and

determining, based on the result of said fine matching, whether or not said non-compressed image matches said compressed image.

10. An image recognition apparatus which detects matching between a non-compressed image and a compressed image, said compressed image being stored in memory as image data which are compressed using orthogonal transform into a two-dimensional spatial frequency region, said compressed image data including, for each of blocks each comprising a predetermined number of pixels in the original uncompressed image, a DC component corresponding to the average intensity of the respective block, comprising:

means for dividing said non-compressed image into one or more blocks corresponding in size to said blocks in said compressed image;

means for computing the average intensity for each of the blocks of said non-compressed image;

means for extracting DC components from said compressed image;

means for performing rough matching between said non-compressed image and said compressed image based on the DC components extracted from said compressed image and the average intensity computed for each of the blocks of said non-compressed image;

means for restoring the compressed image when the degree of matching in the rough matching exceeds a predetermined threshold;

means for performing a fine matching between said non-compressed image and said restored image based on the image data for said restored image and the image data for said non-compressed image; and

means for determining, based on the result of said fine matching, whether or not said non-compressed image matches said compressed image.