Stroke position classification in breast self-examination using parallel neural network and wavelet transform

This study focuses on improving the stroke position classification for the implementation of vision-based breast self-examination guidance system. Previous works have not tackled different variation of breast forms and size and other environment factors. We propose the use of multiple neural networks with parallel computing for more robust classification. Each neural network will be trained for different cases of breast forms and sizes. This creates invariance in breast forms and sizes. Our technique utilized color moments and daubechies-4 wavelet transform for extracting the features in each frames, as the input to the neural networks. This modified approach can classify the stroke position of different breast forms at 89.5% accuracy.