查詢結果分析
來源資料
頁籤選單縮合
題名 | 以輪廓特徵做平面圖形辨識= |
---|---|
作者 | 蔡明俊; 溫永豐; |
期刊 | 金屬工業 |
出版日期 | 19930300 |
卷期 | 27:2 1993.03[民82.03] |
頁次 | 頁78-87 |
分類號 | 312.84 |
語文 | chi |
關鍵詞 | 平面圖形辨識; 輪廓特徵; |
中文摘要 | 本文之目的在於利用影像處理與類神經網路的方法,發展出一套整合的辨識系統,此系統不但能正確地辨認平面的圖形輪廓,且在辨識時,並不受圖形平移、大小、旋轉等因素的影響。在辨識圖形的過程中,擷取圖形的特徵是相當重要的步驟,本文擷取的圖形特徵包括兩部份:第一部份是先將圖形輪廓資料轉換成邊界點與形心間距離的一維統計資料,再取這一維資料的慣量(Moment)作為圖形的特徵;第二部份是偵測圖形的特徵點,將所測得特徵點的數目當作圖形的特徵;由這兩部份所得的特徵向量後,接下來的步驟是進行圖形的判別,由於是將影像數位化的關係,特徵向量的值會有變動,故在進行判別的時候,時而會有模糊不定的情況發生,為了避免此一缺失,本文是利用類神經網路中的逆向傳遞學習網路進行圖形的判別。吾人發現,訓練類神經網路時,若將特徵向量的值先行正規化,則訓練網路的時間戶大量縮短。 |
英文摘要 | The purpose of this paper is to develop an integrated planar shape pattern recognition system. Using the image-processing technique and the method of neural-network, the computer can automatically recognize basic planar shapeto increase the machine vision ability. Futhermore, it can be expanded an automatic identifying system which can recoginze components in a factory. In the processing of shape recognition, the most important step is to extract the features of a shape. The process to adopt the shape-feature includes two parts. The first part is to transfer two-dimensional shape contour data into one-dimensional data which contain the information of the distance between edge-points and center point. The moments of the one-dimensional statistical data were then taken as the shape-features. The second part is to detect the corner-point of the shape. The number of the corner-point was regarded as a shape-feature. Those feature are then combined into a feature-vector which can represents the originsl shape. The next step is to classify the shape. Because of the data were digitized, the values of the feature-vector vary even for the same pattern, In the neural network recognition process, backpropatation method was used to avoid the fuzzy-condition of the feature-vector. It is shown the training time will be dramastic decreased if the value of each feature vector are all normalized. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。