基於視覺注意與多重特徵的快速物件辨識

Abstract

在電腦視覺的領域裡，物件辨識一直是個極具挑戰的問題。物件辨識的困難點在於如何排除外在背景干擾得到圖片中正確的物體，並對物體本身擷取特徵使其成功辨識出物件。常見的特徵抽取方式為對整張圖片做全域的特徵抽取，但全域的特徵抽取易受外在環境干擾，因此本文提出基於視覺注意的快速物件辨識，以醒目性偵測結果輔助物件辨識，藉以降低背景影響，接著以有即時性及可靠性優點的局部二值模式及色彩直方圖做特徵抽取，以達到快速辨識的需求，最後依特徵相似度排序，達成辨識目的。本論文選取Caltech256資料庫來對物件辨識問題進行實驗，並且比較辨識結果的準確率，實驗結果證明醒目性偵測的確能有效提升辨識準確率，且使用多種特徵做相似度計算亦可提高準確率。

Object recognition has been a challenging issue in the field of computer vision. The difficulty of object recognition is how to exclude external background interference and get the object of the picture, then extract the object features and identify the object successfully. The common approach of feature extraction is extracting global feature for the whole picture. However, the global feature extraction tends to be interfered with outside environment. Therefore, we present a fast object recognition which based on visual attention to reduce the background effect. In order to achieve high-speed requirement, we use fast and robust local binary pattern and color histogram to do feature extraction, and then we classify the object according to feature similarity. After sorting the similarities, we achieve the purpose of object recognition. In this paper, we utilize Caltech256 database to experiment on object recognition problems, and we calculate the accuracy rate of our method. The experimental result shows that saliency detection indeed can effectively enhance the accuracy rate of object recognition. Furthermore, make use of a variety of features to make similarity calculation can achieve higher accuracy rate.