Geometric nonlinear diffussion filter based edgemap extraction and its validation of infrared breast images

Autor: C. M. Sujatha, G. Kavitha, J. ThamilSelvi
Rok vydání: 2017
Předmět:
Zdroj: 2017 Third International Conference on Biosignals, Images and Instrumentation (ICBSII).
DOI: 10.1109/icbsii.2017.8082281
Popis: Breast cancer is the leading cancer that affects women in the world. Early diagnosis of cancer prevents morbidity and mortality rate. Thermography is an additional tool for early diagnosis of breast cancer. Low contrast, poor Signal to Noise Ratio (SNR) and complex breast boundaries are the inherent limitation of breast thermal images which makes segmentation a challenging task. In this work, an attempt is made to extract edge map from Type-I and Type-II breast images using Geometric Non linear Diffussion Filter (GNLDF) method to aid accurate segmentation. GNLDF is an iterative nonlinear filter which reduces the noise and preserves the edges based on the local image topology. The extracted edge maps are validated against the input images using betametric and Gradient Magnitude Similarity Deviation (GMSD). Result showed that edge map extracted using GNLDF method is distinct for both types of images. The beta metric and GMSD values are found to be 0.29 and 0.004 indicating high resemblance of filtered image edges with the input images. GNLDF method has improved the SNR of Type-I and Type-II images by 35dB compared to Perona Malik method. Hence, the edge map extracted using GNLD filter shall aid accurate segmentation of breast tissue to identify pathology for clinical interpretation and diagnosis.
Databáze: OpenAIRE