Content Based Image Retrieval using Improved Local Tetra Pattern and Neural Network
Zain Shabbir, I. Arshad, Gulistan Raja, Awais Khan · The Nucleus · 2016
Because of the exponential increase in digital images, image databases have grown to a much large volume that retrieval of required images from these databases is a very difficult task. Image retrieval can also be practiced via human annotation but it cannot be trusted. So, now a days, a more relied and effective method to retrieve relevant images from a large databse is Content Based Image Retrieval (CBIR). These days, main focus is to achieve a more accurate CBIR algorithm so that retrieval efficiency can be increased. In this paper, we have proposed a CBIR algorithm which retrieves images from database with increased precision. For feature extraction, we have used Improved Local Tetra Patterns which is a texture descriptor and based on the direction of pixel. Direction of all the pixels in an image is calculated; based on direction of each pixel, an 8-bit pattern is achieved which is further divided into 7 binary patterns and 1 magnitude pattern. For experimental purposes, we have used Corel database which has been used by most researchers. After performing experiment on the said dataset, improved precision rates were observed when proposed technique was compared with some previous approaches. Average precision observed in our experiment was 82%.