Image processing device and image processing method for processing images using memory-efficient principal component analysis

Patent metadata

Field	Value
Publication number	US-9213914-B2
Application number	US-201313866147-A
Country	US
Kind code	B2
Filing date	Apr 19, 2013
Priority date	Jul 6, 2012
Publication date	Dec 15, 2015
Grant date	Dec 15, 2015

Abstract

A pixel set formation unit in an image analysis unit of an image processing device forms pixel sets from original images subject to analysis. A principal analysis unit of the image analysis unit performs principal component analysis in units of pixel sets. A synthesis unit synthesizes results of analysis in units of pixel sets so as to generate images of eigenvectors of a size of the original images. An image generation unit displays the images of the eigenvectors and stores data for an image generated by using the images of the eigenvectors in a generated image storage unit.

First claim

What is claimed is:

1. An image processing device comprising:
a principal component analysis unit configured to subject a plurality of original images to principal component analysis;
a display unit configured to display images of eigenvectors representing principal components obtained as a result of the principal component analysis; and
an image generation unit configured to acknowledge user input related to adjustment of an image generated by using the images of eigenvectors, and to store the data for the image generated as a result in a storage device or to output the data to a display device,
wherein the principal component analysis unit performs principal component analysis in units of pixel sets produced by retrieving a predetermined number of pixels identically located in the plurality of original images,
the image processing device further comprising a synthesis unit configured to synthesize the eigenvectors to a size of the original images, by returning elements of the eigenvectors obtained for respective pixel sets to positions of the corresponding pixels,
wherein the principal component analysis unit switches between (a) extracting pixels to form the pixel sets from random positions in an image plane of the original images, (b) extracting pixels to form the pixel sets at predetermined intervals in the image plane of the original images, and (c) extracting pixels to form the pixel sets by forming blocks of predetermined size, in accordance with a result of subjecting the original images to frequency analysis,
wherein the pixel sets are formed until each pixel from the original images has been placed in a pixel set.

2. The image processing device according to claim 1 ,
wherein the principal component analysis unit divides an image plane of the original images into blocks of a predetermined size and forms the pixel sets for the respective blocks.

3. An image processing method comprising:
reading data for a plurality of original images from a storage device or an input device and subjecting the data to principal component analysis;
causing a display device to display images of eigenvectors representing principal components, the images obtained as a result of the principal component analysis;
acknowledging user input related to adjustment of an image generated by using the images of eigenvectors, and storing the data for the image generated as a result in a storage device or outputting the data to a display device;
performing principal component analysis in units of pixel sets produced by retrieving a predetermined number of pixels identically located in the plurality of original images,
and synthesizing the eigenvectors to a size of the original images, by returning elements of the eigenvectors obtained for respective pixel sets to positions of the corresponding pixels,
wherein the principal component analysis switches between (a) extracting pixels to form the pixel sets from random positions in an image plane of the original images, (b) extracting pixels to form the pixel sets at predetermined intervals in the image plane of the original images, and (c) extracting pixels to form the pixel sets by forming blocks of predetermined size, in accordance with a result of subjecting the original images to frequency analysis.

4. A non-transitory computer-readable recording medium encoded with a computer program comprising:
a principal component analysis module configured to read data for a plurality of original images from a storage device or an input device and subject the data to principal component analysis;
a display module configured to cause a display device to display images of eigenvectors representing principal components, the images obtained as a result of the principal component analysis; and
a processing module configured to acknowledge user input related to adjustment of an image generated by using the images of eigenvectors, and to store the data for the image generated as a result in a storage device or to output the data to a display device,
wherein the principal component analysis module performs principal component analysis in units of pixel sets produced by retrieving a predetermined number of pixels identically located in the plurality of original images,
the computer program further comprising a synthesis unit configured to synthesize the eigenvectors to a size of the original images, by returning elements of the eigenvectors obtained for respective pixel sets to positions of the corresponding pixels,
wherein the principal component analysis module switches between (a) extracting pixels to form the pixel sets from random positions in an image plane of the original images, (b) extracting pixels to form the pixel sets at predetermined intervals in the image plane of the original images, and (c) extracting pixels to form the pixel sets by forming blocks of predetermined size, in accordance with a result of subjecting the original images to frequency analysis.

Assignees

• Sony Computer Entertainment Inc
• Sony Corp

Inventors

• Oka Masaaki

Classifications

• G06V20/66 Primary

  Trinkets, e.g. shirt buttons or jewellery items (recognising microscopic objects G06V20/69) · CPC title

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• G06F18/2135

  based on approximation criteria, e.g. principal component analysis · CPC title

• G06V10/40

  Extraction of image or video features · CPC title

• G06V10/467

  Encoded features or binary features, e.g. local binary patterns [LBP] · CPC title