Image retrieval using multi texton co-occurrence descriptor

One of many method for image retrieval is Multi Texton Histogram (MTH) that incorporated feature extraction technique. Though the MTH is able to represent the image very well, it’s still has weaknesses. First, the MTH is only using local features to represent image. Second, in the process of pixel pair detection using texton, there is information missing that caused image representation may degrade. This study proposes a new method in order to extract image features for the image retrieval system. The proposed method is named Multi Texton Co-Occurrence Descriptor (MTCD). The MTCD is extracting color, texture and shape features simultaneously using texton, and then calculating image representation globally using Gray Level Co-occurrence Matrix (GLCM). This study used 300 Batik images and 15000 Corel images as datasets. Image similarity is calculated using Canberra and MTCD performance is measured using precision and recall. Our experiments show that by adding 2 new textons and GLCM, the precision rate is increased by 2.86% for Batik dataset, by 3.40%for Corel 5,000 and by 3.06% for Corel 10,000. We conclude that MTCD performance is superior than MTH.