Breast Thermogram Denoising based on Bidimensional Empirical Mode Decomposition and Adaptive Wiener Filter

摘要

Breast thermograms are inherently amorphous in nature and lack with clear edges. They have low signal to noise ratio and are low contrast images. For accurate clinical diagnosis of abnormality, denoising as an essential preprocessing of thermal images is necessary. In this work, an attempt is made to denoise the breast thermograms to improve the clinical analysis in early diagnosis of presence of cancerous tissue. The breast thermograms are recorded using Meditherm IRIS medical thermal system with high array resolution and spectral response under controlled protocol. The captured thermograms are subjected to Anscombe transformation to convert the Poisson distributed date into Gaussian. The transformed images are decomposed into Intrinsic Mode Functions (IMF) by sifting process using Bidimensional Empirical Mode Decomposition (BEMD). First IMF component is subjected to noise removal technique using adaptive Wiener filter. The images are reconstructed by summing the noise free IMF, other IMFs and residue. Denoised breast thermograms are obtained by taking inverse Anscombe transformation of reconstructed image. The results show that BEMD is able to decompose the image into IMFs and its residues. The noise filter appears to be optimum in removing the noise component in homogenous areas compared to edges. Hence, it appears that the denoising technique can be used to improve the accuracy of detecting the presence of tumor in mass screening of breast cancer detection.