A Hybrid Feature Extraction Scheme Based on DWT and Uniform LBP for Digital Mammograms Classification
Sanae Berraho, Mounir Ait Kerroum, Youssef Fakhri · International Review on Computers and Software (IRECOS) · 2015
Breast cancer can be treated most effectively when detected in its early stage. To support radiologists to detect breast abnormalities earlier and more rapidly, creation of a system to analyze automatically mammograms is important. Several techniques can be used to accomplish this task. This paper presents a new approach for digital mammograms classification based on hybrid feature extraction. After decomposing mammogram images in wavelet basis for global feature extraction, block-based uniform local binary pattern is used to extract local features from the resulting sub-bands. Principal component analysis is employed for dimensionality reduction The Support vector machine is then used to construct a supervised classifier. Comparisons with another artificial intelligence algorithm i.e. KNearest Neighbor were also made using two distance calculation methods, namely Euclidian and City-block. Experimental results on Digital Database for Screening Mammography (DDSM) show that the proposed method can classify normal from abnormal mammograms more effectively and has proved also to be better in comparison to other traditional recognition methods.