Kabushiki Kaisha Toshiba

Haruo Asada

Shuichi Tsujimoto

Bunpei Irie

A method and an apparatus for pattern recognition utilizing the multiple similarity method, capable of taking structural features of a pattern to be recognized into account, so that sufficiently accurate pattern recognition can be achieved even when the pattern may involve complicated and diverse variations. The method includes the steps of: counting a number of occurences within each one of localized regions which subdivides a pattern to be recognized, of local patterns indicating possible arrangements of picture elements; deriving a vector quantity indicating distribution of black picture elements which constitute the pattern, from the numbers of occurrences of the local patterns; calculating multiple similarity defined in terms of square of inner product of the vector quantity and one of prescribed standard vectors representing standard patterns; and recognizing the pattern by identifying the pattern with one of the standard pattern whose corresponding standard vectors gives the maximum values for the multiple similarity.

Method and apparatus for a pattern recognition

An image processing apparatus comprises an input image memory for storing input image data obtained by coding original image data with run length codes for a first direction, a reading circuit for sequentially reading from the input image memory the data of every two lines which are adjacent to each other in a second direction orthogonal to the first direction, an exclusive OR processing circuit for carrying out an exclusive OR operation on the data of the every two lines read by the reading circuit to detect second directional endpoint positions of the original image data without decoding the input image data into the original image data, and a second directional processing circuit for carrying out a predetermined image process for the second direction according to the first directional endpoint positions detected by the exclusive OR processing circuit.

Image processing apparatus

A image inclination detecting method and an apparatus achieving the method calculate distances between leading edges or between trailing edges of runs (consecutive white or black pixels) that are located on adjacent two scanning lines respectively. From occurrence frequencies of these edge distances, a representative value that may indicate an inclination of the image most clearly is found, and, based on the representative value, the inclination of the image is computed.

Growing community of inventors

Yokohama, Japan

Haruo Asada