Semi automatic segmentation of breast thermograms using variational level set method

In this work, an attempt has been made to delineate the breast tissue from breast thermograms. Breast thermograms used for this experimentation and validation were taken from public database PROENG. Firstly, a spatially adaptive wavelet based denoising method was used to improve signal to noise ratio. Secondly, contrast-limited adaptive histogram equalization method was used to improve the contrast. Finally, an active contour based level set method using variational formulation without reinitialization was adopted to segment the region of interest. The segmented results were validated and evaluated against the ground truth taken from the database by calculating few overlap measures considering pixels in the region of interest. Further, the segmented results of preprocessed image were compared with the results obtained on raw images. The results demonstrated the accurate segmentation of region of interest using variational level set method. Better edges were observed by improving signal to noise ratio and contrast of the image. Consequently, subjective and experimental analysis shows that the removal of noise and improvement in contrast led to precise segmentation of region of interest.