원문정보
초록
영어
To overcome the problem in recognition of retina texture with noise, an improved multi scale retina texture recognition algorithm is proposed based on fusion area threshold. Integrated with regional threshold filtering, it is useful to suppress the effect of non-uniform blocks for texture recognition. Also, the nonlinear multi-scale transform can analyze the geometric space coefficient of retinal vessels with multi direction invariant features. The maximum likelihood local mean standard deviation analysis is used for texture parameters estimation and recognition. Accurate identification of texture feature is obtained with the noise reduced greatly. The simulation results show that the algorithm can well characterize the retinal vascular texture, with good performance in different texture feature recognition. With good robustness, the recognition accuracy is improved also.
목차
1. Introduction
2. Basic Knowledge and Algorithm Principle
2.1. Non Sub-Sampled Contourlet Transform
2.2. High Order Cumulant Slice Texture Threshold Equalization Cancellation Algorithm
3. Multi-Scale Recognition Based on Non-Subsampled Contourlet Transform
4. Simulation and Result Performance Analysis
5. Conclusions
References