Improved CBIR System Using Multilayer CNN

The knowledge of computer vision and machine learning are used in the Content Based Image Retrieval (CBIR) system. Due to the revolution of the internet and availability of 4G internet speed with low cost, lots of images are being generated and shared on a day which required a good and accurate retrieval system. The textual system creates language indistinctness problem. So, CBIR system is established to overcome the problem like manual annotation and language indistinctness. The knowledge of computer vision is used for image retrieval based on some low-level features like color, shape, and texture. The most important parts of the CBIR system are the computational complexity and the retrieval accuracy. The machine learning helps to improve the accuracy of the system. In this paper, Convolution Neural Network (CNN) is used due to the property of direct feature extraction from the image. Computer vision techniques are used to extract low-level features from the image while CNN is used to extract high-level features from the image. The CNN extract the features from the image as well as to classify the image which removes unnecessary images and increase retrieval accuracy. As the number of convolution layers increases the more high-level features can be extracted but it will increase the training time. The results of CBIR are compared for one layer of CNN and two layers of CNN. The average precision rate is used to compare CNN based technique with the different fusion techniques of computer vision.