Abstract

ABSTRACTSpeckle noise is a principal and unavoidable source of visual degradation in several real world images obtained from coherent imaging systems such as ultrasound, SAR, and laser. Over the last few decades, several mathematical models have been used to study despeckling and have influenced policy. This paper review and compare the partial differential equation-based non-linear diffusion models for despeckling of digital grey scale natural and real images, available in the literature. First, we present a detailed study of different classes of partial differential equation-based models namely, logarithmic transform, anisotropic diffusion, and total variation based models, which have different assumptions about noise removal and explore the effect of treatments. Then, to test the effectiveness of different discussed classes of PDE-based models, a set of natural images with different geometric regions, three real ultrasound images and two real SAR images are used. Also, to assess the quality of considered test images after despeckling using numeric metrics, three with-reference and two without-reference indexes are used in the experimental study. A few experiments with natural and real images reveal that the filters based on logarithmic transformation are cost effective with low noise suppression capability. Whereas, the anisotropic diffusion and total variation-based approaches show the good filtering performance with high computational cost. Therefore, rigorous qualitative and quantitative studies confirm that the seemingly subtle variation in model assumptions can have remarkable impact on despeckling.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call