一般平面図形の識別のための特徴量
スポンサーリンク
概要
- 論文の詳細を見る
What feature properties should be extracted is the most important problem in the recognition of non-symbolic patterns or so-called 'general shapes'. In the case of the recognition of general shapes, the categories are determined by the purpose the recognizer has in mind, although symbolic patterns like alphabets have their categories fixed a priori. Therefore, the problem of pattern recognition of general shapes cannot be dealt with in a general way.The authors discuss this problem from the following points of view;1) The recognizer determines the'searching shapes'as the category set according to the purpose in his mind.2) The primary measure in the recognition of the general shapes is'similarity'.In this paper, the authors propose a new method for selecting feature properties in the recognition of the general shapes. The method is based on the human impression of similarity:1) Enumerate as many feature properties as possible, after determining the searching shapes.2) For every searching shape, generate the so-called random shapes having the same properties by using the Pattern Reproduction Method.3) Select such limited number of properties among the candidate properties that can sufficiently generate shapes similar to the searching shapes.Verification for this method has been carried out on a digital computer, dealing with Jordan curves as sampled patterns to discuss 23 kinds of psychological properties of the shape. As a result, four properties out of the 23 are selected to be significant for the recognition of both convex and concave shapes.
- 公益社団法人 計測自動制御学会の論文
公益社団法人 計測自動制御学会 | 論文
- Self-Excited Oscillation of Relay-Type Sampled-Data Feedback Control System
- タイトル無し
- Mold Level Control for a Continuous Casting Machine Using an Electrode-Type Mold-Level Detector
- Assessment and Control of Noise:Pollution by Noise from General Sources
- Information network system and home automation.