Information Processing Method and Apparatus, and Image Pickup Device

• US 20070244842A1
• Filed: 06/02/2005
• Published: 10/18/2007
• Est. Priority Date: 06/03/2004
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Mitarai, Yusuke, Matsugu, Masakazu, Mori, Katsuhiko, Ishii, Mie
• Granted Patent: US 7,991,719 B2
• Time in Patent Office: Days
• US Class Current: 706/18
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/082 modifying the architecture, e.g. adding, deleting or silencing nodes or connections
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]

Abstract

An output value of neuron within an objective layer of a hierarchical neural network is computed. The data of the output value of neuron is stored in a memory only if the output value of neuron is greater than or equal to a predetermined value by referring to the computed output value of neuron within the objective layer. When the data of the output value of neuron on a former layer of objective layer is read from the memory, the data having a predetermined value is read, instead of the data of the output value of neuron not stored in the memory.

Claims

1. An information processing apparatus for detecting a predetermined pattern in an input image by a process employing a hierarchical neural network in which a plurality of detection layers and integration layers are arranged alternately, the detection layer detecting and outputting one or more feature amounts from the output of the integration layer at the former stage and the integration layer integrating the outputted features from the detection layer at the former stage and outputting the result of integration, comprising:

output value computation means for computing an output value of neuron within an objective layer, using an output value of neuron within a former layer of the objective layer and a weight coefficient between the objective layer and the former layer, sequentially by setting each layer as the objective layer;

first storage control means for storing data of the output value, which is greater than or equal to a predetermined value, in a memory by referring to the output value of each neuron within the detection layer computed by said output value computation means;

second storage control means for storing data of the output value of each neuron within the integration layer computed by said output value computation means, in the memory; and

supply means for supplying the data of the output value of neuron within the former layer of the objective layer to said output value computation means by reading the data of the output value from the memory and supplying the data having a predetermined value instead of the data of the output value of neuron not stored in the memory.

2. The information processing apparatus according to claim 1, wherein said output value computation means omits the computation of an output value of neuron, when all the data to be referred to in computing the output value of neuron within the detection layer is not supplied by said supply means.

4. An image pickup device mounting the information processing apparatus according to claim 1.

3. An information processing apparatus for modifying a weight coefficient between layers to obtain a desired output in a hierarchical neural network in which a plurality of detection layers and integration layers are arranged alternately, the detection layer detecting and outputting one or more feature amounts from the output of the integration layer at the former stage and the integration layer integrating the outputted features from the detection layer at the former stage and outputting the result of integration, comprising:

output value computation means for computing an output value of neuron within an objective layer, using an output value of neuron within a former layer of the objective layer and a weight coefficient between the objective layer and the former layer of the objective layer;

storage control means for storing data of the output value, which is greater than or equal to a predetermined value, in a memory by referring to the output value of each neuron within the processing object layer computed by said output value computation means;

modification means for modifying the weight coefficient between the objective layer and the former layer of the objective layer, based on an error between the data of the output value of the objective layer consisting of the data of the output value stored in the memory and the data having a predetermined value instead of the data of the output value of neuron not stored in the memory and the data of the desired output; and

supply means for supplying the data stored in the memory by said storage control means, which is referred to by said output value computation means to compute the output value of each neuron within the objective layer, to said output value computation means by reading the data of the output value from the memory.

5. An information processing method for detecting a predetermined pattern in an input image by making a computation process employing a hierarchical neural network in which a plurality of detection layers and integration layers are arranged alternately, the detection layer detecting and outputting one or more feature amounts from the output of the integration layer at the former stage and the integration layer integrating the outputted features from the detection layer at the former stage and outputting the result of integration, comprising:

an output value computation step of computing an output value of neuron within an objective layer, using an output value of neuron within a former layer of the objective layer and a weight coefficient between the objective layer and the former layer of the objective layer, sequentially with setting each layer as the objective layer;

a first storage control step of storing data of the output value, which is greater than or equal to a predetermined value, in a memory by referring to the output value of each neuron within the detection layer computed in the output value computation step;

a second storage control step of storing data of the output value of each neuron within the integration layer computed in the output value computation step, in the memory; and

a supply step of supplying the data of the output value of neuron within the former layer of the objective layer, which is referred to in the output value computation means, by reading the data of the output value from the memory and supplying the data having a predetermined value instead of the data of the output value of neuron not stored in the memory.

7. A program for causing the computer to perform the information processing method according to claim 5.

8. A computer readable storage medium storing the program according to claim 7.

6. An information processing method for modifying a weight coefficient between layers to obtain a desired output in a hierarchical neural network in which a plurality of detection layers and integration layers are arranged alternately, the detection layer detecting and outputting one or more feature amounts from the output of the integration layer at the former stage and the integration layer integrating the outputted features from the detection layer at the former stage and outputting the result of integration, comprising:

an output value computation step of computing an output value of neuron within an objective layer, using an output value of neuron within a former layer of the objective layer and a weight coefficient between the objective layer and the former layer of the objective layer;

a storage control step of storing data of the output value, which is greater than or equal to a predetermined value, in a memory by referring to the output value of each neuron within the processing object layer computed in the output value computation step;

a modification step of modifying the weight coefficient between the objective layer and the former layer of the objective layer, based on an error between the data of the output value of the objective layer consisting of the data of the output value stored in the memory and the data having a predetermined value instead of the data of the output value of neuron not stored in the memory and the data of the desired output; and

supply means for supplying the data stored in the memory in the storage control step, which is referred to in the output value computation step to compute the output value of each neuron within the objective layer, to the output value computation step by reading the data of the output value from the memory.

9. A program for causing the computer to perform the information processing method according to claim 6.