適用於行動裝置的物件辨識系統

在數位影像如此發達的時代，利用影像來查詢資訊已經是一件非常方便的事情，可是要如何快速找出我們要的物件已成為各專家研究的議題。在過去，許多研究只能使用特定的角度來做影像查詢，可是不同的使用者在拍攝時會有不同的取景觀點，要如何快速找出我們要的物件成為了本論文想要解決的問題。影像查詢中包含特徵擷取和特徵比對兩個步驟，在特徵擷取部分，我們加入了影像的顏色資訊，因為過去許多研究往往只採納灰階或是二元影像，並不足以表現更豐富的特徵資訊，本文做法為將其色彩空間先做轉換到人類視覺較敏銳的HSV色彩空間，接者將HSV色彩空間資訊以紋理特徵法取得特徵；另外並將HSV色彩資訊量化至72色，這樣即可達到在行動裝置上快速的辨識物件。在特徵比對方面，本研究為了要讓使用者可以根據自己對於拍攝影像具有不同的觀點，先將物件定位並找出重心，再利用重心轉換到座標的圓心上，如此一來，即可在比對時達到抵抗旋轉的方式。實驗中，本研究將拍攝到的圖片分別做位移(左與右)、縮小與旋轉，最後再透過Euclidean distance作為物件之間特徵相似度的依據，根據上述四種變形的情況，其準確率皆為100%；相似率根據上述四種變形後，其準確率皆為100%。故本方法對於不同角度、大小及方位均有極高之辨識準確率。

關鍵字

影像查詢；物件辨識

並列摘要

In this times, digital image is developed quickly, use digital image to retrieval what you want. It can be a very convenient thing. But how can we find out what exact object rapidly become a research topic. In the past, a majority of research just can retrieve one direction. Nevertheless, different users have different views. In this thesis, how to find out exact object rapidly turn into a problem which I want to solve. There are two characteristics in image retrieval. Which is feature extract and feature comparison. In the feature extract part, we added color feature information. Because there was lots of researchers only adopted gray image or binary image. It can’t represent more detail information about image. So I transform RGB color space to HSV color space first. After that, the texture is extracted. At same time, I quantized HSV to 72 colors numeric. It can use in object recognized over the mobile. In the feature comparison part, in order to let every users can take photos by their views. I find out the object center of gravity first. Then, transform the centroid into polar coordinates. In this way, it can solve image rotation problem in this section. In experimental, this research will transform original photos into shift (left and right), shrink and rotation. At last, I used Euclidean distance to represent objects similarity. The precision and similarity are both 100% for above deformation. Eventually, this research has high precision in different angles, sizes and directions.

並列關鍵字

Image retrieval ； Object recognition

參考文獻

[18] C. Palm, Color texture classification by integrative co-occurrence matrices, Pattern Recognition 37 (5) (2004) pp.965–976.

[19] G.H. Liu, L. Zhang, et al., Image retrieval based on multi-texton histogram,

Pattern Recognition 43 (7) (2010) pp.2380–2389

[20] G.H. Liu, J.Y. Yang, Image retrieval based on the texton co-occurrence matrix,

Pattern Recognition 41 (12) (2008) pp.3521–3527

國際替代計量

適用於行動裝置的物件辨識系統

未授權

主題瀏覽