| Summary: | Image segmentation is an important step to suitable extraction of features of objects from images. However, the presence of noise interferes in segmentation quality; for example, by generating the detection of false edges (or false borders). To diminish the problems caused by the presence of noise in images, various smoothing techniques have been proposed to pre-process the original images. Those methods reduce the noise presented in input images, but they can also strongly affect the borders of the objects, leading to the loss of important details, such as the original roughness of the contours or the elimination of the borders of small objects. Among the existing smoothing techniques, one of the most promising is based on the use of anisotropic diffusion, which allows a selective smoothing that decreases the undesirable effects caused by noise presented in the input image and preserves the edges of the objects. However, the success of this smoothing method is strongly reliant on the number of iterations performed that depends on the input image. In this work, we propose the use of derivative filters for the definition of the appropriate number of iterations adopted by the smoothing method based on anisotropic diffusion, when it is applied for the removal of noise usually present in images of skin lesions. The experimental results demonstrate that the developed solution is promising, being able to determine the adequate number of iterations for smoothing the input images avoiding the excessive loss of details of the borders of the lesions presented in images.
|
|---|