Image Scene Classification based on Latent Semantic Analysis

Digital image classification is becoming increasingly important. A digital image can be represented by its low-level features. Semantic analysis is a common technique for scene image classification. How to reduce the semantic gap between the high-level semantic and low-level features is a significant problem. This paper proposes a novel scene image classification method. Latent Semantic Analysis (LSA) is applied to scale feature dimensions, delete noise, and select the important latent semantic features of each scene. To increase classification accuracy, low-level features should contain diversity. The proposed mechanism is applied to classify images by measure the scene similarity of each image. Experimental results demonstrate that the proposed mechanism can classify image scenes successfully and has a better correct classification ratio than other analytical methods. The proposed image classification method reduces the semantic gap and close to human semantics.