ANALYSIS ON FINE-GRAINED IMAGE RECOGNITION

Abstract

In addition to being able to distinguish between a dog and any bird, humans are amazingly capable of visualising or picturing fine-grained detail, we can also discern an American Bulldog from an English Bulldog To educate or train machines to apprehended in a fine-grained manner, fine-grained image recognition was brought to the academic community. FGIR is used in different academic and business settings. applications in both industry and academia. These applications have had a positive impact on fields like conservation and commerce. The objective of FGIR has been a long-term goal of the deep learning community. It is to retrieve and identify images from several intra-class categories of a super-class category, such as various dogs, aircraft, and plant species, retail product types, etc. Determining how to distinguish between items that are remarkably similar in appearance but different in fine-grained attributes is, consequently, the main difficulty. Intricate nuances and subtle visual signals that might not be clearly seen in low resolution photos are frequently used in fine-grained image identification. We have put efforts into an approach to this problem using super-resolution to recover finer details, such as textures, patterns, or minor differentiating traits. These details are essential for the categorization or localization of fine-grained categories. Usually, accurate localization of objects or particular regions of interest within an image is necessary, which is performed by specialised feature detection. We also performed a survey on recent trends in the field, and we tried to identify and classify the generalised techniques used in FGIR into groups based on the types of techniques followed and the results that we were trying to achieve.