Content-based image retrieval through semantic image segmentation

The term Content-based image retrieval (CBIR) is widely used to describe the process of retrieving desired images from a large database on the basis of features that can be automatically extracted from the images themselves. Earlier CBIR methods were based on extracting the low-level features of the image like shape, color, texture etc. But such systems lacked efficiency because the image concepts were not correctly identified. So there is need for an intelligent system to identify the concepts correctly. Semantic segmentation methods help in this regard by analyzing an image region wise. The introduction of deep learning techniques have brought about a significant performance improvement in semantic image segmentation methods. In this paper DeepLab v3 is used for semantic segmentation and ResNet-34 is used for further classification and image retrieval. Segmentation is performed on PASCAL VOC 2012 dataset, the regions extracted from this is further used for classification. This method provides superior results to other baseline methods.