Adaptive Noise Reduction and Edge Preservation Using Anisotropic Diffusion Kuwahara Filtering on Traffic Images

Abstract

This paper proposes a comprehensive method to improve the segmentation of traffic images by leveraging anisotropic diffusion filtering and Two-Directional Two-Dimension Principal Component Analysis (2D2PCA). Anisotropic diffusion filtering is first applied to the traffic images to preserve critical edges and structures while effectively reducing noise. Despite these benefits, the filtered images may still contain artifacts that can compromise segmentation accuracy for tasks such as object detection and lane delineation. To address this, we introduce a novel post-processing technique designed to refine segmentation results, enhancing the boundaries of segmented objects and eliminating spurious artifacts. The proposed methodology involves multiple phases: Noise Reduction, Edge Preservation, Improved Segmentation, Enhanced Visibility, and Adaptive Filtering. Each phase employs distinct algorithms to perform specific operations on transportation images. Our approach integrates the edge-preserving capabilities of anisotropic diffusion filtering with the dimensionality reduction power of 2D2PCA, achieving over a 70% reduction in original image size while maintaining boundary edges in the reduced images. Experimental evaluations on various traffic scenes demonstrate the effectiveness of our method, showing significant improvements in segmentation accuracy compared to traditional methods. These reduced-size images are highly beneficial for future work in traffic image analysis tasks.