| 作成者 |
|
|
|
|
|
| 本文言語 |
|
| 出版者 |
|
| 発行日 |
|
| 収録物名 |
|
| 巻 |
|
| 号 |
|
| 開始ページ |
|
| 終了ページ |
|
| 出版タイプ |
|
| アクセス権 |
|
| JaLC DOI |
|
| 関連DOI |
|
|
|
|
|
| 関連URI |
|
|
|
|
|
| 関連情報 |
|
|
|
|
|
| 概要 |
仮説検証型の画像理解システムについて、従来の視覚システムの問題点を明らかにし、初期視覚の特徴抽出においてもモデル駆動型の予測や注意集中が必要である事を述べ、この点からニューラルネットワークを用いた初期視覚システムのモデル化を行った。本モデルにより任意のスケールで安定した輪郭線が抽出できることを示すと主に、高レベルからの局所的な注視制御が可能になることを明らかにした。
In image underst...anding, it is necessary to combine data-driven analysis and mode1-driven prediction adequately. Therefore recently, the approaches of hypothesis-verification style come to be active. In these approaches, there are two main difficulties: how to control the adaptation of hypothesis and how to speed up large hypothesis verification processes. For these difficulties, the approach based on neural networks, which are performed in asynchronous massive parallel, is seemed to be useful. In early vision, there are few approaches from the above view point. In this paper, we, first, discuss that the model-driven prediction is also necessary for early vision. And, then, we propose a neural network model for early vision. The proposed network is constructed with three sub-networks based on Hopfield’s network. The first one is for regularization with line process. The second one is for classification. The line process detect edges in the classified image, and is performed well on arbitrary scale in the regularization. The last sub-network enables higher level processes to control the scale locally. Finally, we present results of our experiments of edge detection based on the above method.続きを見る
|
Mendeley出力
