Image-based Multi-resolution Analysis Of Plant Leaves Identification Systems

Plant is a familiar living form with the largest population and the widest distribution on the earth,which plays an important role in improving the environment of human beings.It is very important to recognize the plant species correctly and quickly in the collecting and preserving genetic resources,the discovery of new species,plant resource surveys and plant species database management.This thesis proposes an implementation frame of plant recognition based on leaves' images by means of extracting texture features.Multi-resolution Analysis are widely applied in image processing field,especially in texture analysis.Hence,we applied it to image-based leaves recognition.We adopt Gabor filters and Wavelet Transform which is quite popular in recent years in plant images analysis.A newly appeared means of texture analysis called local binary pattern is introduced to make the images rotation invariant and eliminate the impact of illumination.It described the texture and structural characteristics by depicting the gray level changes in the pixels' neighborhood.In this thesis a method of block Local Gabor Binary mode is proposed in order to overcome the defect caused by LBP for its exponential growth in storage space.It makes full use of the multi-scale characteristics of Gabor filters and the constant binary encoding characteristics of LBP which is gray-scaled and rotation invariant.It can reduce the effects of illumination.At the same time,the block processing enhances the ability of local characterization.Therefore,it makes them more suitable for image processing. After extracting the texture features,we utilize support vector machine as classifier to recognize the plant leaves.The experimental results show that it is an effective method.At the same time,we have developed a plant leaf image recognition system using Visual C++ 6.0. Now there are 27 classes which contain about 500 plant leaves having been successfully classified.Finally,the works of this thesis are briefly summarized and viewed,and further research works are also discussed and proposed.