公开(公告)号：JP2001339565A

公开(公告)日：2001-12-07

申请号：JP2000159354

申请日：2000-05-30

Applicant: IBM

Inventor： NAKAMURA YOJI

IPC: G06T1/00 ,  H04N1/00 ,  H04N1/40

Abstract: PROBLEM TO BE SOLVED: To simply and quickly detect an abnormal image read. SOLUTION: Image data information (data size or compression rate) after compression is compared with a reference value decided in cross-reference with an abnormal image read in advance to discriminate the presence/absence of the abnormal image read. In the case of using the data size, the read fault is discriminated when the data size after compression is smaller than a predetermined reference value. Moreover, when the data compression rate is used the compression rate is greater than the predetermined reference value, the occurrence of the read fault is discriminated. According to the method, the abnormal image read can simply and quickly be discriminated.

公开(公告)号：JPH01277981A

公开(公告)日：1989-11-08

申请号：JP10446988

申请日：1988-04-28

Applicant: IBM

Inventor： NAKAMURA YOJI ,  RICHIYAADO JII KEESHI ,  TOYOKAWA KAZUHARU

IPC: G06K9/70 ,  G06K9/68

Abstract: PURPOSE: To reduce a storage area and samples required for correction by storing the images of the characters of a document in an image buffer, storing the codes and flags of the characters of high and low recognition in a result buffer and performing the correction by a learning function. CONSTITUTION: A document scanning device 1 scans the document and stores the character images for one page in the image buffer 2. A character segmentation circuit 3 segments all the characters of one row and stores them in the frame of an operation buffer 4. A control circuit 5 operates a logic circuit 10, takes out character recognition trees 6A-6E and all the character codes from a tree memory 6, recognizes the character images and stores a result in an auxiliary buffer. The character code is stored in the result buffer 7 when the accuracy of recognition is higher than a prescribed value and the character code and the flag are stored when it is lower. The image and character code of an erroneously recognized character are stored in a learning buffer 9 and the learning of the character recognition tree, that is the correction, is performed.