An Object Query System using HMMD and Texture Features for Photos of Taiwan Wild Birds

Abstract

This research attempts to propose an image query system based on the color and texture features to allow users to find more in-depth understanding of ecological knowledge by means of “querying photos by a photo”. The frequencies of color structure descriptor based on HMMD color space is adopted as the color feature of the subjects. On the other hand, gray level co-occurrence matrix and Haar wavelet transform are used as the texture feature. For the decision making, four types of feature matching are compared: the Euclidean distance, the Cosine similarity, the Neural Network and Decision tree. The experiment is carried out using the photos of wild birds in Taiwan. We choose 15 bird species which are easier to collect their photos, with each of them 15 photos being the training sample. The test results showed that using only the color features allows the system achieved top3 correct rate of more than 95% accuracy. However, using gray-level co-occurrence matrix as texture characteristics cannot serve as an effective algorithm. In the future, experiments using more bird species will be conducted and, different texture features will be surveyed and tested in order to enhance the system performance.