An LBP-HOG Descriptor Based on Matrix Projection For Mammogram Classification
This work addresses computational efficiency in mammogram classification, but it is incremental as it adapts existing descriptors with matrix projections.
The authors tackled the high computational cost of iterative scanning in image feature descriptors by proposing matrix multiplication-based LBP and HOG variants, achieving promising classification accuracy and efficiency on a mammogram dataset.
In image based feature descriptor design, local information from image patches are extracted using iterative scanning operations which cause high computational costs. In order to avoid such scanning operations, we present matrix multiplication based local feature descriptors, namely a Matrix projection based Local Binary Pattern (M-LBP) descriptor and a Matrix projection based Histogram of Oriented Gradients (M-HOG) descriptor. Additionally, an integrated formulation of M-LBP and M-HOG (M-LBP-HOG) is also proposed to perform the two descriptors together in a single step. The proposed descriptors are evaluated using a publicly available mammogram database. The results show promising performances in terms of classification accuracy and computational efficiency.