ABSTRACT

This paper presents a low dimensional edge descriptor, based on local diagonal and non-diagonal maxima-minima pattern for medical images. The methods utilize local relationship of a middle pixel to its diagonal/non-diagonal maxima-minima. The scheme reduces the length of feature vector to a great extent without compromising the quality of edge map. The designed methods are validated on digital X-ray for bone fracture detection, dental images and ultrasound images. Improved depth local binary pattern, anisotropic diffusion & Canny edge detection, Canny and Sobel methods are also used for a comparative analysis. The local diagonal maxima-minima pattern and local non-diagonal maxima-minima pattern methods are robust to noise, while preserving the useful edge information. The experimental results reveal the superiority of proposed methods as compared to the state-of-the-art methods in terms of accuracy, Jaccard similarity index, specificity, sensitivity and dice similarity coefficient.

摘要

本文提出了一种基于医学图像的局部对角线和非对角线极大极小模式的低维边缘描述符。该方法利用中间像素与其对角线/非对角线最大值-最小值的局部关系。该方案在不影响边缘图质量的情况下，很大程度上减少了特征向量的长度。设计的方法在用于骨折检测的数字 X 射线、牙科图像和超声图像上进行了验证。改进的深度局部二值模式、各向异性扩散和 Canny 边缘检测、Canny 和 Sobel 方法也用于比较分析。局部对角极大极小模式和局部非对角极大极小模式方法对噪声具有鲁棒性，同时保留有用的边缘信息。