Image Retrieval using Modified Multi Texton and Rotation Invariant Local Binary Pattern

Image retrieval is a standard method used to search a digital image in large image databases. One of the most critical things in image retrieval is the feature extraction process. Several techniques can be used in the image feature extraction process, one of which is Multi Texton Histogram (MTH). MTH is usually carried out with two main processes, namely color quantization, and texture orientation. Improvement of MTH method performance can be made by adding texton. In our experiment, we test the method by using rotated images. In this paper, we develop the use of an invariant method for rotation, namely the Local Binary Pattern Rotation Invariant (LBPROT) method to improve image retrieval precision. The results of combining these two methods produce features of each image. The features produced by each image are compared using a distance matrix. A query image that has the smallest distance matrix value is an image that has the same class as the compared database image. Based on our experiments on typical images, the use of Modified MTH and LBPROT can improve image retrieval performance by increasing the percentage of precision and recall by 3.63% and 0.62%. However, The Modified MTH is better when compared to the merging of the Modified MTH and LBPROT on 45 °, 135 °, 225 °, and 315 ° angle rotations with precision values by 24% -27%. Meanwhile, our testing result using rotated images, the combination of Modified MTH and LBPROT produces an increase in precision at angles rotation of 90 °, 180 °, and 270 ° by 2% -3%.