ニューラルネットによる手書き機械図面の自動認識 : 第2報,多段階NNに基づくベクトルデータの作成

概要

論文の詳細を見る
The key step in drawing recognition is the extraction and separation of drawing features, such as lines and intersections. In this paper, we propose a multilevel neural network (MLNN) procedure for extracting these features. Even though the first-level neural network (NNLE) requires only a few training patterns to learn the distinctions between line segments, it has strong capability, high accuracy and speed of detecting line elements under many difficult conditions. The second-level neural network (NNIE) can further recognize intersections in the remaining image data. In this stage, the obtained information includes not only the positions of intersections, but also the directions of the line segments in these intersections. We also developed a new and simple method that can facilitate the process of determining vectors. Finally, all the results are reconstructed as an overall view of the drawings. Several examples have shown the success of the MLNN system in recognizing handwritten mechanical drawings.
一般社団法人日本機械学会の論文
1996-06-25