A Flower Image Classification Based On Constraint Sparse Coding And Feature Weighted Fusion

With the rapid development of digital technology,the digital images show explosive growth.Faced with such a wide variety of images,it is an urgent and important problem for effectively improving its classification effect.In particular,flower species are numerous.Furthermore,each kind of flower images exist their own characteristics such as flower body's shape and colour.Moreover,different kinds of flower images have the same commonalities such as green leaf and background,and classification effect of flower is not ideal.Therefore,it has an important research value how to utilize the image classification technology to exactly identify flower categories and help people to be familiar with the appearance and its growth environment of flowers in terms of flower categories,and accordingly improving the flower cultivation technology.Sparse coding is an important method for coordinately representing specific components and common components of features.In order to effectively improve the discriminability of images categories,a flower image classification method based on constraint sparse coding and feature weighted fusion in this paper is studied.It mainly includes the following three aspects:(1)A visual dictionary generation model based on constraint sparse coding is presented.This generation model aims at generating a category-specific visual dictionary for each kind of image features which only belongs to corresponding category,and generating a shared visual dictionary for all kinds of image features which belongs to all the categories.In order to improve the discriminability of feature representation and dictionary,two constraint terms in the extensible expression of sparse coding are firstly added.The first one is used to restrict the influence of other kinds of category-specific dictionaries on the representation of this kind of image features.The second one is used to restrict the coherence between various atoms in the dictionaries.Then a parameter of penalty term corresponding to global dictionary is split into two parameters of penalty term corresponding to category-specific dictionaries and shared dictionary.These two parameters are used to respectively control the sparsity ofcoefficient parts,which correspond to these two types of dictionaries in the global coding.Finally experiments are conducted in the Oxford Flower-17 dataset,and the effectiveness of this dictionary generation model is validated.(2)A sparse coding classification method based on feature weighted fusion is presented.According to the classification thought of sparse coding,a weighted coding classification model in the basis of global coding classification model and local coding classification model is firstly presented.This classification model imposes fusion in reconstruction losses in a weighted way,which respectively correspond to global coding and local coding.Then,after introducing the extraction of multiple kinds of image features,a sparse coding classification method based on feature weighted fusion is presented.This classification method again imposes the weighted fusion in the reconstruction losses of these multiple kinds of features after a fusion.The final reconstruction loss is used to identify the image categories.Finally,experiments are conducted in the Oxford Flower-17 dataset,and the effectiveness of this image classification method is validated.(3)A prototype system of flower image classification is achieved.According to the above research achievements,a flower image classification prototype system based on constraint sparse coding and feature weighted fusion in the Matlab 7.0 platform is designed and achieved.