Current Proceedings on Technology

Yazarlar: Emre Akarslan, Rifat Edizkan

Konular:-

Anahtar Kelimeler:Local Binary Pattern,Local Phase Quantization,Object Classification

Özet: Image representation is a very important topic in visual object classification. The success of the image representation affects the success of the classification system directly. The appearance of an interest region can be well characterized by the distribution of its local features. Local features are widely used for object classification in recent years. Local Binary Pattern (LBP) and Local Phase Quantization (LPQ) are two well-known operators for shape and texture descriptor. The LBP and LPQ give illumination- and blur-insensitive description of an image respectively. In the LBP operator, neighbors of a center pixel are tresholded by center pixel. The LPQ operator quantizes the Fourier transform phase in local neighborhoods. In this study we purpose the use of two local features, Local Binary Pattern and Local Phase Quantization, for the representation of objects to improve the classification accuracy in object classification task. LBP and LPQ histograms are extracted from the overlapping regions in an image. The histograms are concatenated into an enhanced feature vector as an object descriptor. Bag of words (BoW) model is used for constructing visual vocabulary. The objects are classified by using linear SVM classifier method. In the experiments, LBP, LPQ and the combination of LBP and LPQ features are evaluated in the object classification task separately and the accuracies are compared. The experiments on Caltech-4 database show that the combining of the LBP and LPQ features improves the classification accuracy.

---