臺灣博碩士論文加值系統

到目前為止,許多作者已經提出各種不同不變矩量,來辨識幾何失真下的圖
型,但這些方法只針對在有位移,大小,旋轉失真下的物體才有效.在本論文
裡,先前的正規化演算法可以轉換線性扭曲失真物體到標準位置,使它不變
於扭曲失真.經由實驗結果和理論證明這些方法無法同時正規化旋轉對稱
物體和非旋轉對稱物體.因此沒有任何正規化方法可以適合實際應用,我們
提供三種模型來克服這些困難.同時,組合解扭曲方法和旋轉不變矩量可以
建構一組扭曲不變矩量的完全集.實驗結果證明這組扭曲不變矩量集合的
性能比以前的方法優良.最後正規化方法直接應用到圖型的邊界使的計算
複雜度大大的減少.

So far many authors have proposed various moment invariants for
the recognition of the patterns under geometric distortion, but
they are invariant only with respect to translation, scaling and
rotation. In this thesis, existing normalization algorithm can
transform the affine-distorted patterns to the standard position
which is invariant under affine distortion. Our experimental
results and theory show they can not simutaneouly normalize
rotationally symmetric shapes and non-rotationally symmetric
shapes. Therefore, no normalization method can fit the real
application. We proposed three models which can overcome the
above difficulty in real application. In the meantime, combining
de-skewing algorithm with rotation invariants can construct a
complete set of affine moment invariants.The experimental
results show that the performance of the set of affine moment
invariants is superior to the performance of the existing one.
Finally, normalization method is directly applied to the contour
of the shape so that its computation complexity is tremendously
reduced.