	US 7,483,565 B2
	Image processing device and method, learning device and method, recording medium, and program
Tetsujiro Kondo, Tokyo (Japan); Takashi Sawao, Tokyo (Japan); Junichi Ishibashi, Saitama (Japan); Takahiro Nagano, Kanagawa (Japan); Naoki Fujiwara, Tokyo (Japan); Toru Miyake, Tokyo (Japan); and Seiji Wada, Kanagawa (Japan)
Assigned to Sony Corporation, Tokyo (Japan)
Filed on Mar. 07, 2008, as Appl. No. 12/44,557.
Application 12/044557 is a division of application No. 10/545074, previously published as PCT/JP2004/001607, filed on Feb. 13, 2004.
Claims priority of application No. 2003-050208 (JP), filed on Feb. 27, 2003.
Prior Publication US 2008/0199072 A1, Aug. 21, 2008
Int. Cl. G06K 9/62 (2006.01); G06K 9/36 (2006.01); G06K 9/46 (2006.01)

U.S. Cl. 382—155 [382/238; 382/190; 348/E11.002; 348/E7.004; 348/E7.016]

3 Claims

1. A learning device for learning predicting means which predict, from input image data made up of multiple pieces of pixel data obtained by an imaging device having a plurality of pixels each having time integration effects or spatial integration effects imaging light signals of the real world, high-quality image data with higher quality than said input image, said learning device comprising:

image data continuity detecting means for detecting image data continuity corresponding to the continuity of light signals of said real world, for a plurality of first perimeter pixels within said input image data that correspond to a pixel of interest within said high-quality image data;

first extracting means for extracting a plurality of second perimeter pixels within said input image data that correspond to said pixel of interest in said high-quality image data, based on said image data continuity detected by said image data continuity detecting means;

second extracting means for extracting a plurality of third perimeter pixels within said input image data that correspond to said pixel of interest, based on said image data continuity detected by said image data continuity detecting means;

features detecting means for detecting features of said plurality of second perimeter pixels extracted by said first extracting means; and

learning means for learning said predicting means for predicting said pixel of interest from said plurality of third perimeter pixels extracted by said second extracting means for each of said features detected by said features detecting means.