Image Scene Classification Based on Fisher Discriminative Analysis and Sparse Coding

Abstract

Bag of visual words (BoVW) is currently widely utilized as an image representation model. However, conventional BoVW methods usually bring drawbacks such as large representation errors, lack of spatial information and weak discrimination. In order to overcome those, this paper proposes a new image scene classification algorithm based on fisher discriminative analysis and sparse coding. Firstly, we construct the non-negative sparse locally linear coding to encode the local features with neighbors' visual vocabularies, thus to make effective use of images' spatial information. Secondly, we add fisher discriminative analysis to construct a non-negative sparse locally linear coding model based on fisher discriminative criterion constraint, thus to obtain the images' discriminative sparse representation, and promote the spatial separability of sparse coefficients and enforce the classification capability of images' sparse representation. Finally, we combine SVM classifier to perform scene classification. Experiment results show that our algorithm efficiently utilizes spatial information of images and incline to seek images' discrimination representations, and improves the classification performance, thus it's more suitable for image classification.