PProCRC: Probabilistic Collaboration of Image Patches

TL;DR

PProCRC introduces a probabilistic framework for collaborative image patch representation that inherently manages background and outliers, offering a non-iterative solution that improves fine-grained species recognition accuracy.

Contribution

It proposes a novel probabilistic method for collaborative image patch analysis that eliminates pre-processing and provides a closed-form solution, outperforming existing CRC approaches.

Findings

Outperforms previous CRC methods on three datasets

Effective with multiple CNN backbones

Provides a non-iterative, closed-form solution

Abstract

We present a conditional probabilistic framework for collaborative representation of image patches. It incorporates background compensation and outlier patch suppression into the main formulation itself, thus doing away with the need for pre-processing steps to handle the same. A closed form non-iterative solution of the cost function is derived. The proposed method (PProCRC) outperforms earlier CRC formulations: patch based (PCRC, GP-CRC) as well as the state-of-the-art probabilistic (ProCRC and EProCRC) on three fine-grained species recognition datasets (Oxford Flowers, Oxford-IIIT Pets and CUB Birds) using two CNN backbones (Vgg-19 and ResNet-50).