INCOHERENT DICTIONARY LEARNING WITH LOCALITY CONSTRAINED LOW-RANK REPRESENTATION FOR IMAGE CLASSIFICATION

Abstract

Low-rank representation (LRR) plays a significant role in image classification tasks due to its ability to capture the underlying structure and variations in image data. However, traditional low-rank representation-based dictionary learning methods struggle to leverage discriminative information effectively. To tackle this issue, we propose an incoherent dictionary learning approach with locality-constrained low-rank representation (LCLRR-IDL) for image classification. Firstly, we introduce low-rank representation to handle potential data contamination in both training and test sets. Secondly, we integrate a locality constraint to recognize the intrinsic structure of the training data, ensuring similar samples have similar representations. Thirdly, we develop a compact incoherent dictionary with local constraints in the low-rank representation to classify images, even in the presence of corruption. Experimental results on public databases validate the effectiveness of our approach.