Hybrid Approach for Content-Based Image Retrieval using VGG16 Layered Architecture and SVM: An Application of Deep Learning

Abstract

Advancements in the sector of computer and multimedia technology and introduction of the World Wide Web have increased the volume of image databases and collections, for example medical imageries, digital libraries, art galleries which in total contain millions of images. The retrieval process of images from such huge database by traditional methods such as Text Based Image Retrieval, Color Histogram and Chi Square Distance may take a lot of time to get the desired images. It is necessity to develop an effective image retrieval system which can handle these huge amounts of images at once. The main purpose is to build a robust system that builds, executes and responds to data in an efficient manner. A Content-Based Image Retrieval (CBIR) system has been developed as an efficient image retrieval tool where user can provide their query to the system to allow it to retrieve user's desired image from the image collection. Moreover, the emergence of web development and transmission networks and also the number of images which are available to users continue to grow. We propose an effective deep learning framework based on Convolution Neural Networks (CNN) and Support Vector Machine (SVM) for fast image retrieval. Proposed architecture extracts features using CNN and classification using SVM. The results demonstrate the robustness of the system.

An oversampling method for multi-class imbalanced data based on composite weights

“…this reason, we used the composite indicators such as Accuracy, F-value and G-mean to evaluate the performance of classification algorithms as a whole in imbalanced data [40,41]. 1.…”

Attention Module Magnetic Flux Leakage Linked Deep Residual Network for Pipeline In-Line Inspection

“…The VGG16 network is a deep network model developed by the Computer Vision Team of Oxford University and researchers of Google DeepMind in 2014. The network has 16 training parameters, with a simple network structure and excellent generalization performance when transferred to other image data [ 28 , 29 ]. The VGG16 model exhibited excellent performance in classification and achieved the best effect on multiple datasets at that time.…”

Deep Learning-Based Image Extraction

“…The first two layers include the identical padded and 64 channel having 3*3 filter size. Alongside a max pool layer of stride (2, 2), two layers comprise fully connected levels of 128 sampling frequency and step size (3,3). A max pooling stride (2, 2) which is the same as the layer preceding it comes next.…”

“…After that, there are different pairs of 3 fully connected layers, and then a maxpooling layer comes next. Every filter is having the similar padding and has 512 filters of size (3,3). The stacks of 2 convolution layers then receives this image.…”

“…We provide a CBIR strategy for detecting objects of absorption in a collection of documents, while CNN's being used to get the portrayal of the recoverable picture in order to go around such problems. [3] The two major components of a CBIR project execution are extracting features and comparable assessment, and researchers are working on these for a very lengthy period. There are many solutions offered, but these remain the most challenging issues in the ongoing CBIR research because of the discrepancy between the high image pixels captured by machines and the lower quality pixels captured by amazing human perceptions.…”