HESSIAN-LoG : A NOVEL DOT ENHANCEMENT FILTER

In order to make regions appearing as dots on a chest radiograph more clearly in visual, a novel filtering method, Hessian-LoG filter, is developed in this paper. The proposed filter consists of two major operations. First, using the eigenvalues of Hessian matrix detects dot points in chest radiography.The derivatives of a Hessian matrix at multiple scales are convolved with an input image at each pixel.To determine the local shape of structures at pixels in each scale, the eigenvalues of Hessian matrix are analyzed at each pixel in the resulting image.The diameter of the detected dot is equal to the kernel size with the highest classifier value.Finally, dot images are convolved with the Laplacian of Gaussian (LoG) operator.Experiments show that the proposed enhancement filter can simultaneously enhance dot-like objects and suppress line-like structures, and thus improve the sensitivity of nodule detection.