Image processing apparatus and feature detection method

• US 9,710,877 B2
• Filed: 06/19/2015
• Issued: 07/18/2017
• Est. Priority Date: 12/28/2012
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Koezuka, Tetsuo, Nagato, Tsuyoshi
• Primary Examiner(s): Conner, Sean
• Assistant Examiner(s): PATEL, PINALBEN V
• Application Number:
  US14/744,607
• Publication Number:
  US 20150287160A1
• Time in Patent Office: 760 Days
• CPC Class Codes: G06F 18/2111 by using evolutionary computational techniques, e.g. genetic algorithms
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06T 1/20 Processor architectures; Processor configuration, e.g. pipelining
• CPC Class Codes: G06T 2207/20072 Graph-based image processing
• CPC Class Codes: G06T 2207/30164 Workpiece; Machine component
• CPC Class Codes: G06T 3/00 Geometric image transformations in the plane of the image
• CPC Class Codes: G06T 7/0004 Industrial image inspection
• CPC Class Codes: G06V 10/451 with interaction between the filter responses, e.g. cortical complex cells
• CPC Class Codes: G06V 10/457 by analysing connectivity, e.g. edge linking, connected component analysis or slices
• CPC Class Codes: G06V 10/771 Feature selection, e.g. selecting representative features from a multi-dimensional feature space

Abstract

An image processing apparatus is disclosed. A processor performs a process that arranges multiple image filters in a tree structure and acquires features of multiple input images captured in different environments. A storage part stores a filter table that maintains information concerning the multiple image filters. In the process, an image filter having a different type for each of layers, in which multiple image processes are hierarchized, is selected from the filter table. A population, which includes the multiple image filters in the tree structure, is generated.

Claims

1. An image processing apparatus, comprising:

a processor configured to perform a process that arranges multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing and acquires features of multiple input images captured in different environments; and

a storage part configured to store a filter table that maintains the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property,

wherein the process includes

selecting, by referring to the filter table, one or more image filters within a depth restricted to a layer for each of layers;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.

2. The image processing apparatus as claimed in claim 1, wherein the generating of the population generates the population, in which the multiple image filters are arranged in the tree structure, by using genetic programming.

3. The image processing apparatus as claimed in claim 2, wherein the generating of the population arranges the multiple image filters in the tree structure in which a target image is set as a root, and selects the input images to an end node in a predetermined layer.

4. A feature detection method performed by a computer, the method comprising:

performing, by the computer, a feature detection process for arranging multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing, and acquiring features of multiple input images captured in different environments,

wherein the feature detection process includes

selecting one or more image filters within a depth restricted to a layer for each of layers by referring to a filter table stored in a storage part, the filter table maintaining the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.

5. A non-transitory computer-readable recording medium storing a program which, when executed by a computer, causes the computer to perform a feature detection process for arranging multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing, and acquiring features of multiple input images captured in different environments, the feature detection process comprising:

selecting one or more image filters within a depth restricted to a layer for each of layers by referring to a filter table stored in a storage part, the filter table maintaining the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.

6. An image processing apparatus for executing a process program, in which multiple filters are formed in a tree structure, with respect to multiple input images captured by an imaging device, the apparatus comprising:

a processor configured to perform a process that arranges multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing and acquires features of multiple input images captured in different environments; and

a storage part configured to store a filter table that maintains the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property,

wherein the process includes

classifying the multiple input images into clusters based on feature amounts concerning the multiple images at an initial learning stage;

selecting a representative image close to a center of the feature amounts for each of the clusters; and

generating the process program through learning employing genetic programming by using learning data, which include multiple representative images selected for the clusters,

wherein the generating of the process program further includes

selecting, by referring to the filter table, one or more image filters within a depth restricted to a layer for each of layers;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.

7. The image processing apparatus as claimed in claim 6, wherein the feature amounts are acquired based on an average luminance, a luminance histogram, and space frequency information for each of the multiple images.

8. The image processing apparatus as claimed in claim 6, wherein during a learning process using the genetic programming, the feature amounts are calculated by the process program of the tree structure in middle of generation, and the learning data is re-selected.

9. The image processing apparatus as claimed in claim 8, wherein the average luminance is acquired as the feature amount from each of the filters forming the process program of the tree structure.

10. The image processing apparatus as claimed in claim 6, wherein clustering is performed based on the feature amounts with respect to the learning data, representative samples are selected respectively from sample groups, and the learning data is updated.

11. An image processing method performed in a computer, the method comprising:

performing, by the computer, a feature detection process for arranging multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing, and acquiring features of multiple input images captured in different environments,

wherein the feature detection process includes

classifying the multiple input images into clusters based on feature amounts concerning the multiple images at an initial learning stage;

selecting a representative image close to a center of the feature amounts for each of the clusters; and

generating the process program through learning employing genetic programming by using learning data, which include multiple representative images selected for the clusters,

wherein the generating of the process program further includes

selecting one or more image filters within a depth restricted to a layer for each of layers by referring to a filter table stored in a storage part, the filter table maintaining the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.

12. A non-transitory computer-readable recording medium storing a program which, when executed by a computer, causes the computer to perform a feature detection process for arranging multiple image filters in a tree structure hierarchized by an order of an image emphasis process, a threshold process, and a binary image processing, and acquiring features of multiple input images captured in different environments, the feature detection process comprising:

classifying the multiple input images into clusters based on feature amounts concerning the multiple images at an initial learning stage;

selecting a representative image close to a center of the feature amounts for each of the clusters; and

generating the process program through learning employing genetic programming by using learning data, which include multiple representative images selected for the clusters,

wherein the generating of the process program further includes

selecting one or more image filters within a depth restricted to a layer for each of layers by referring to a filter table stored in a storage part, the filter table maintaining the multiple image filters, each of which is associated with a property indicating one of the image emphasis process, the threshold process, and the binary image processing, and a depth of a node restricted by the property;

generating a population that includes the multiple image filters being selected in the tree structure; and

conducting an evolutionary process using genetic programming, in which a crossover process is conducted with respect to the population based on a first restriction rule of a crossover pair, which restricts selection of the crossover pair for each of the layers, and a mutation process is conducted with respect to the population based on a second restriction rule of mutation, which restricts types of the mutation for each of the layers.