Image Classification Based On Feature Coding

In the field of computer vision,image classification is a basic work.It is also a critical step,directly influencing many other computer vision tasks,and plays a major role.In this paper,we focus on the classification task of the medium scale image database.Most people make use of visual dictionary model to solve the problem.The reason of the model's popularity is that each part is independent and flexible.Especially in recent years,feature coding is more critical than other parts.The feature coding method has been continuously improved,thus the accuracy of categorization was improved a lot.But the image classification task still face many challenges: the accuracy shows poorly in intra-class variation.The main reason lies in that most feature coding methods only consider coding in the feature space,and proposed the importance of locality,saliency and similarity constraint regardless of spatial domain of the image.Meanwhile,we found that the typical feature coding method has the instability of coding vectors in our practical experiments.Thus,our paper's theme is feature coding and research many methods about this theme.In this paper,In order to solve the above problems,this paper proposes a stable local feature coding method for image classification based on spatial context,and the main contents are as follows::1.Firstly,a lot of research on the technology of feature coding is carried out,and the theory and principle of methods based on global feature and local feature coding are described in detail.At the same time,their shortcoming and advantages and the relationship between the two methods is also discussed.2.Secondly,our paper investigate each part in the bags of words model and compare relevant part.Then,we choose the best performance method in our project.3.Lastly,this paper makes use of useful information from feature space and spatial domain of image to code.We find adjacent space of each feature point quickly and establish a judgment space context model to enhance the discriminative power of feature coding;and it takes nonnegative definitions constraint to improve stability of coding vectors.On the premise of maintaining local feature coding's locality,smooth sparsity and reconstruction accuracy,we increase the spatiality and the non negativity and stability of coding vectors...