System and method for image recognition

• US 6,185,337 B1
• Filed: 12/12/1997
• Issued: 02/06/2001
• Est. Priority Date: 12/17/1996
• Status: Expired due to Term

• Current Assignee: Honda Giken Kogyo Kabushiki Kaisha (Honda Motor Company)
• Original Assignee: Honda Giken Kogyo Kabushiki Kaisha (Honda Motor Company)
• Inventors: Koerner, Edgar, Masutani, Tomohiko, Tsujino, Hiroshi
• Primary Examiner(s): Au, Amelia
• Assistant Examiner(s): MILLER, MARTIN E
• Application Number:
  US08/989,416
• Time in Patent Office: 1,152 Days
• Field of Search: 382/118, 382/206, 382/224, 382/226, 382/227, 382/160
• US Class Current: 382/227
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

In an artificial intelligence system for image recognition, a global image of an object is input from a camera or other optical pick-up device, and is processed in a global image processing means, which performs analytical processing on the global image by extracting global characteristics of the input image and evaluating consistency of the extracted global characteristics. Simultaneously, the image data is processed in a local image processing means which undertakes analytical processing on a plurality of local images defining local portions of the image to be recognized. The local image processing means is constructed by plural modules, each further defined by sub-modules, which conduct respective analyses corresponding to local images having characteristics useful in recognizing the global image, wherein each local processor extracts characteristics of an input local image and evaluates consistency of the extracted characteristic with the object to be recognized. Importantly, the global image processing means receives inputs from the local modules, and deactivates functions of local modules which are inconsistent with the global characteristics, while activating and promoting functions of local modules which are consistent with the global characteristics. Through top-down control from the global image processor, as well as inter-module signals between respective local processing modules, since inconsistent processes are quickly discovered,

Claims

1. An image recognition system for recognizing an object using a global image processing means and a local image processing means performing parallel processing on a global image of an object of recognition input from a camera or other optical pick-up device and on characteristic local images, comprising:

global image processing means for performing analytical processing on a global image, comprising means for extracting a global characteristic of an input global image and means for evaluating consistency of said extracted global characteristic;

local image processing means for performing analytical processing on a plurality of local images, said local image processing means being made up from a plurality of local modules corresponding respectively to each local image, each of said local modules comprising means for extracting a characteristic of an input local image and means for evaluating consistency of the extracted characteristic with the image to be recognized,

wherein said global image processing means receives an input image and exchanges signals with said local modules, and deactivates functions of said local modules which are inconsistent with said global characteristic, while activating functions of said local modules which are consistent with said global characteristic, thereby effecting a hierarchical control wherein consistency evaluations made in said global image processing means influence activation or deactivation of functions in said local modules, and wherein respective local modules of said local image processing means control functions of said global image processing means based on signals indicating consistency with said local images.

2. The image recognition system according to claim 1, wherein each of said local modules making up said local image processing means comprises plural local modules corresponding to different topologies of the same local image, and determined by local images which correspond in location with each other.

3. The image recognition system according to claim 1, wherein each of said local modules making up said local image processing means comprises plural local modules corresponding to different topologies of the same local image, and determined by local images which correspond in form with each other.

4. The image recognition system according to claim 1, wherein each of said local modules making up said local image processing means comprises:

a first sub-module comprising means for extracting a characteristic of an input local image;

a second sub-module having stored therein knowledge of an eigenspace of a given local icon of a pre-learned object of recognition; and

a third sub-module comprising means for evaluating consistency based on a projected distance from a result of projecting the image information of said input local image having the characteristic extracted by said first sub-module onto an eigenspace stored in said second sub-module.

5. The image recognition system according to claim 4, wherein said first sub-module has stored therein knowledge of an average vector of the given local image of said pre-learned object of recognition.

6. An image recognition system for recognizing an object from a global image of the object of recognition input from a camera or other optical pick-up device, and from characteristic local images, comprising:

global image processing means for performing analytical processing on a global image, comprising means for extracting a global characteristic of an input global image and means for evaluating consistency of said extracted global characteristic;

local image processing means for performing analytical processing on a plurality of local images, said local image processing means being made up from a plurality of local modules arranged in correspondence with a topology of a local image, each of said local modules comprising (a) a first sub-module comprising means for extracting a characteristic of an input local image, (b) a second sub-module having stored therein knowledge of an eigenspace of a given local icon of a pre-learned object of recognition, and (c) a third sub-module comprising means for evaluating consistency based on a projected distance from a result of projecting the image information of said input local image having the characteristic extracted by said first sub-module onto an eigenspace stored in said second sub-module,

wherein said global image processing means receives both an input image and inputs from said local modules, and deactivates functions of said local modules which are inconsistent with said global characteristic while activating functions of said local modules which are consistent with said global characteristic, and

wherein among plural local modules which have input thereto image information from neighboring regions and which are in charge of different local images, one local module having an activated first sub-module therein outputs a signal for deactivating the first sub-module of another local module in charge of a different local image.

8. An image recognition system according to claim 6, wherein plural local modules in charge of the same local image are arranged in correspondence with a topology of image information formed by the local image.

10. An image recognition system according to claim 6, wherein when consistency is confirmed based on an evaluation in said third sub-module of one of said local modules, the first sub-module of the same local module is deactivated.

7. An image recognition system for recognizing an object from a global image of the object of recognition input from a camera or other optical pick-up device, and from characteristic local images, comprising:

global image processing means for performing analytical processing on a global image, comprising means for extracting a global characteristic of an input global image and means for evaluating consistency of said extracted global characteristic;

local image processing means for performing analytical processing on a plurality of local images, said local image processing means being made up from a plurality of local modules arranged in correspondence with a topology of a local image, each of said local modules comprising (a) a first sub-module comprising means for extracting a characteristic of an input local image, (b) a second sub-module having stored therein knowledge of an eigenspace of a given local icon of a pre-learned objection of recognition, and (c) a third sub-module comprising means for evaluating consistency based on a projected distance from a result of projecting the image information of said input local image having the characteristic extracted by said first sub-module onto an eigenspace stored in said second sub-module,

wherein said global image processing means receives both an input image and inputs from said local modules, and deactivates functions of said local modules which are inconsistent with said global characteristic while activating functions of said local modules which are consistent with said global characteristic, and

wherein image information is input to said second sub-module of one of said local modules only when said first sub-modules is activated above a predetermined threshold.

9. An image recognition system according to claim 7, wherein plural local modules in charge of the same local image are arranged in correspondence with a topology of image information formed by the local image.

11. An image recognition system according to claim 7, wherein when consistency is confirmed based on an evaluation in said third sub-module of one of said local modules, the first sub-module of the same local module is deactivated.

12. An image recognition system for recognizing an object using a global image processing means and a local image processing means performing parallel processing on a global image of an object of recognition input from a camera or other optical pick-up device and on characteristic local images, comprising:

global image processing means for performing analytical processing on a global image, comprising means for extracting a global characteristic of an input global image and means for evaluating consistency of said extracted global characteristic;

local image processing means for performing analytical processing on a plurality of local images, said local image processing means being made up from a plurality of local modules arranged in correspondence with a topology of a local image, each of said local modules comprising (a) a first sub-module comprising means for extracting a characteristic of an input local image, (b) a second sub-module having stored therein knowledge of an eigenspace of a given local icon of a pre-learned object of recognition, and (c) a third sub-module comprising means for evaluating consistency based on a projected distance from a result of projecting the image information of said input local image having the characteristic extracted by said first sub-module onto an eigenspace stored in said second sub-module,

wherein said global image processing means receives an input image and exchanges signals with the sub-modules of said local modules, and deactivates functions of the sub-modules of said local modules which are inconsistent with said global characteristic, while activating functions of the sub-modules of said local modules which are consistent with said global characteristic, thereby effecting a hierarchical control wherein consistency evaluations made in said global image processing means influence activation or deactivation of functions in said sub-modules of said local modules, and wherein respective sub-modules of said local modules of the local image processing means mutually deactivate functions of the sub-modules of other local modules.

13. An image recognition system for recognizing an object using a global image processing means and a local image processing means performing parallel processing on a global image of an object of recognition input from a camera or other optical pick-up device and on characteristic local images, comprising:

global image processing means for performing analytical processing on a global image, comprising means for extracting a global characteristic of an input global image and means for evaluating consistency of said extracted global characteristic;

local image processing means for performing analytical processing on a plurality of local images, said local image processing means being made up from a plurality of local modules arranged in correspondence with a topology of a local image, and wherein plural local modules in charge of the same local image are arranged in correspondence with a topology of image information formed by the local image, each of said local modules comprising (a) a first sub-module comprising means for extracting a characteristic of an input local image, (b) a second sub-module having stored therein knowledge of an eigenspace of a given local icon of a pre-learned object of recognition, and (c) a third sub-module comprising means for evaluating consistency based on a projected distance from a result of projecting the image information of said input local image having the characteristic extracted by said first sub-module onto an eigenspace stored in said second sub-module,

wherein said global image processing means receives an input image and exchanges signals with the sub-modules of said local modules, and deactivates functions of the sub-modules of said local modules which are inconsistent with said global characteristic, while activating functions of the sub-modules of said local modules which are consistent with said global characteristic, thereby effecting a hierarchical control wherein consistency evaluations made in said global image processing means influence activation or deactivation of functions in said sub-modules of said local modules, and wherein respective sub-modules of said local modules of the local image processing means mutually deactivate functions of the sub-modules of other local modules.