A Review on Fine Grained Categorization of an Image using Part Proposals

Abstract

The intent of the image categorization process is to classify the digital image into one of the classes. General image categorization is comparatively easier than fine grained image categorization but it may fail to discriminate objects belonging to same class like birds, cars, plants etc. Fine grained image categorization needs to emphasize on the tiny details that helps to discriminate between similar objects. Many researchers used object /part based methods under strong supervision and weak supervision. The aim is to generate image representation which can be suitable for fine grained categorization. In the new system, object proposals are extracted from input image. From each object proposal, multi-scale part proposals are generated, from which many useful part proposals are selected. A global image representation is generated using selected useful part proposals. The global image representation is then used to train the classifier for image categorization. Application areas are forestry, agriculture, industry and research societies.