Adaptive Multiscale Edge-Preserving Filtering for Improved Segmentation and Feature in Digital Image Processing

The edge preservation of digital images is a significant issue in digital image processing, especially in activities of segmentation, feature extraction, and structural interpretation. Yet, traditional smoothing and denoising filters are not always able to preserve fine edges or to adapt to spatially varying noise, leading to blurred edges and loss of important image information. The paper presents the adaptive multiscale edge-preserving filtering model that aims to improve the quality of segmentation and accuracy of feature extraction in various imaging scenarios. The approach combines a pyramidal decomposition, hierarchical with spatially adaptive weighting functions to control the degree of smoothing depending on the magnitude of the local gradients, variation of the texture and estimation of noise. A hybrid edge-consistency constraint is used to preserve high profile structural boundaries across scales, whilst smaller variations are effectively flattened. It is experimentally evaluated on natural and medical imaging data and shows much better results in localizing boundaries, noise resilience and precision of the segmentations than classical bilateral, anisotropic diffusion and guided filtering methods. The framework proposed gives an efficient, flexible, and edge-sensitive improvement scheme that can be applied to the computer vision pipeline in the present times.