梯度計算裝置及方法、邊緣偵測系統及方法

專利類型

發明

專利國別 (專利申請國家)

中華民國

專利申請案號

099145523

專利證號

I 444925

專利獲證名稱

梯度計算裝置及方法、邊緣偵測系統及方法

專利所屬機關 (申請機關)

國立中興大學

獲證日期

2014/07/11

技術說明

影像中像素的梯度，能提供大量有關邊界與物件輪廓之資訊；影像梯度為影像中相鄰像素間的灰階強度之差異度。當邊界或物件輪廓受到雜訊汙染，或受到太過暗或太過亮的光線照射時，邊界與物件輪廓常會模糊不清；尤其當兩個同一類型物件有部分重疊時，其重疊部分上的物件輪廓往往是不明顯的，且不易被Roberts Cross operation、Prewitt operation、Sobel operation等傳統梯度法所偵測出。本發明利用像素間的垂直和水平方向之二階導數的組合，來求解任一個像素的最大梯度與其正交方向之對應梯度，並且將此結合結果來表示該像素之最佳梯度量。此技術能有效地將邊界像素之梯度給予加強，且將非邊界像素之梯度給以壓抑，藉此來偵測出模糊的邊界與物件輪廓。 The gradients of an image can provide much information for edges. Gradient describes the grayscale variances of adjacent pixels in an image. If edges or object contours are polluted by noise or too dark or bright light is shined onto an image, the edges and object contours are often indefinite. Especially if two objects with similar type are partially overlapped, the object contours on the overlapped part are often unobvious, and traditional gradient methods, i.e. Roberts Cross, Prewitt, and Sobel operations, cannot effectively detect them. This invention defines the gradient of a pixel as the directional maximum of the contrast function. In the gradient computation, two extreme values attained in orthogonal direction, coinciding with the eigenvalues of a 2×2 Euclidean metric matrix, are used to define the gradient of a pixel. It can effectively strengthen the pixel gradients located at the vicinity of indefinite edges and object contours, and suppress the gradients of non-edge pixels.

備註

本部(文號1090000062)同意該校108年12月31日興產字第1084300814號函申請終止維護專利(中興)

連絡單位 (專責單位/部門名稱)

技術授權中心

連絡電話

04-22851811

網址