A predictive coding approach on microvessel identification via Single-Opponent signals

In this paper, we propose a predictive coding approach to identify microvessel subimages using Single-Opponent (SO) signal. First, we simulate the signal from SO neurons in the positive red and negative cyan channel to segment microvessel regions. Second, the SO signal is convolved with a Laplacian of Gaussian (LoG) filter and then processed in a predictive coding and biased competition (PC/BC) neural network. The texture patterns from SO signal is learnt and represented by the PC/BC model. Finally, we employ Support Vector Machine (SVM) with Spatial Pyramid Matching (SPM) scheme to classify the microvessel subimages on the learnt texture patterns. We have carried out extensive experiments on the identification of microvessel subimages. The proposed method is about 10% superior in F1-measure compared with direct classification on subimages in RGB channel.