Detection and Inpainting of Specular Reflection in Colposcopic Images with Exemplar-based Method

Cervical cancer is the most common genital malignant tumor which seriously threats to women's health, especially in poor areas. As one of the screening methods for cervical lesions, colposcopy plays an important role in the early stage of cervical lesions. However, due to the physiological mucus in the human body, the specular reflection region (SR region) is often shown in a high-definition image obtained by doctors through a colposcopy. The expression of the highlight area in the colposcopic image has a certain similarity with the abnormal metaplasia epithelium (AW area, i.e. the lesion area), which leads to a series of problems in the medical image processing in the later stage, such as the segmentation of the lesion area, misdiagnosis by doctors, etc. Therefore, this paper proposed the exemplar-based inpainting method combining with the improved threshold method to detect the SR regions. In the experiment, 150 sets of data (50 pieces for different degrees of cervical lesions) were inpainted by this paper and the other two existing inpainting methods, and the results were compared with subjective and objective quality. The experiment showed that the exemplar-based method has better effect in removing the specular reflection in cervical images.