• P. Sasikumar

Image retrieval (IR) methods gains more popularity because of its huge accessibility of multi-media datasets. The current IR system had an incredible performance over labeled dataset. But often, data labelling is expensive and infrequently not possible. Thus, unsupervised and self-supervised learning techniques are renowned techniques. Many self or unsupervised techniques are sensitive to the several classes and cannot combine labeled data on availability. With this study, this study focuses on the design of CBIR model using VGG19 Network with Class Attention Mechanism (CBIR-VGGCAM). The CBIR-VGGCAM technique derives a class attention layer (CAL) for capturing the discriminatory class-specific feature. In addition, the Canberra distance is applied as a similarity measurement metric for determining the resemblance among the feature vector of the images. Upon providing the query image (QI), the feature vectors are derived by the VGG19 network and a similarity measurement is made with the feature vector of the image database. Eventually, the images with high resemblance are retrieved in the database. A wide-ranging experiment was take place on benchmark databases and the obtained outcomes depict the advancement of the CBIR-VGGCAM algorithm over the recent approaches.

Content based image retrieval, Class attention layer, VGG19 networks, Deep learning, Corel10K dataset

"Improved Deep Transfer Learning Based VGG19 Network Enabled Content Based Image Retrieval Model", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.9, Issue 9, page no.e55-e63, September-2022, Available :http://www.jetir.org/papers/JETIR2209407.pdf

"Improved Deep Transfer Learning Based VGG19 Network Enabled Content Based Image Retrieval Model", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.9, Issue 9, page no. ppe55-e63, September-2022, Available at : http://www.jetir.org/papers/JETIR2209407.pdf