Universal learning machine and application to english type‐letter recognition
Hisashi Suzuki, Suguru Arimoto · Electronics and Communications in Japan (Part III Fundamental Electronic Science) · 1989
Abstract This paper proposes a universal learning machine (ULM) that absorbs information from environments and acquires an ability to process information professionally according to the environments. Additionally, it applies this information to the recognition of the English type‐letters that are highdensely printed in magazines, books, etc. A window is applied on a black and white image gathered through an image scanner so that only a letter and a head part of the succeeding letter can be observed. Appropriately fixing a distance that measures the similarity between the window images, the recognizer utilizing the ULM searches for some one of the most similar window images to the current from among the window images memorized in the past. Then the recognizer loads the records accompanying it to display for a user the reading of the designated letter and the boundary between this letter and the succeeding letter. the user commands whether the recognizer should revise the memory or continue the recognition process. If the command is for revision, the content of revision is memorized in the form of a binary tree while part of the content memorized in the past is deleted to prevent overflow in system memory capacity. A theoretical evaluation of the performance of the ULM and an experimental evaluation on the English type‐letter recognizer actually constructed are performed.