Thermography is a useful imaging tool using infrared for the early diagnosis of breast cancer. Screening cancer aims to outstrip prognosis by seeing the precancerous stage to give a prominent prescription. Early diagnosis is essential to avoid the fatality rate in abnormal cases. In this article, a novel approach is proposed using image analysis and machine learning techniques. In the present work, thermal images were collected from the visual laboratory. In the pre-processing stage, the contrast of the image is improved by combining top-hat and bottom-hat transforms. The ROI extraction method is the preliminary process to select the right and left breast region and remove the neck and armpit region. Then, the imperfection in the structure of the image has been eliminated by using morphological operations. Statistical, geometrical, and intensity features are extracted from the pre-processed and segmented images. Texture features using a Gray-Level Co-Occurrence matrix are obtained both in the spatial domain and curvelet domain. The curvelet transform is used in the feature extraction stage, and this can be used to find an explanation of the curve discontinuity. The curvelet wrapping is applied, followed by the application of GLCM to extract texture features. In the proposed method, 16 features are used for the automated classification of input thermal images. Different machine learning techniques are explored, and the cubic SVM renders the highest accuracy of 93.3%. A combination of statistical, intensity, geometry features, and texture features extracted from curvelet coefficients provides the highest accuracy.

热成像是使用红外技术对乳腺癌进行早期诊断的有用成像工具。筛查癌症的目的是通过看到癌前期给出明显的处方来超越预后。早期诊断对于避免异常情况下的死亡率至关重要。在本文中，提出了一种使用图像分析和机器学习技术的新颖方法。在目前的工作中，热图像是从视觉实验室收集的。在预处理阶段，通过组合顶帽和底帽变换来改善图像的对比度。ROI提取方法是选择左右乳房区域并去除颈部和腋窝区域的初步过程。然后，通过使用形态学运算消除了图像结构中的缺陷。统计的，几何的 从预处理和分割的图像中提取强度特征。在空间域和曲波域中均获得使用灰度共生矩阵的纹理特征。Curvelet变换用于特征提取阶段，可用于找到曲线不连续性的解释。应用Curvelet包裹，然后应用GLCM提取纹理特征。在提出的方法中，将16个特征用于输入热图像的自动分类。探索了不同的机器学习技术，立方支持向量机提供了93.3％的最高准确性。从Curvelet系数提取的统计，强度，几何特征和纹理特征的组合提供了最高的准确性。