Research And Application Of Improved Gene Expression Programming Based Image Processing Algorithm

Combined genetic algorithms and genetic programming, gene expression programming (GEP) is a new evolutionary computation (EC) method, which has been successfully applied in many fields such as date mining and function finding. Multi-expression programming (MEP) is one of GEP whose chromosome contains multiple expressions. Compared with GEP, MEP can make the population evolution generation of complex problem be well reduced.This thesis mainly focuses on the algorithms improvement of GEP and MEP and their applications in image registration and image retrieval, which are two key research topics in the image processing. By utilizing GEP and MEP, the obtained models of the image registration and image retrieval were all get higher accuracy. The main work and results of this thesis are listed in the following:1. Some basics were introduced in the beginning and at the same time the encoding characteristic of GEP and MEP, summaried the application of GEP in the image processing, proposed the method of image registration based on GEP and image retrieval based on MEP.2. As one application case, the related image registration problem can be transformed to mining function if utilizing GEP’s ability in function funding. On the basis of establishing image registration model according to evolutionary modeling methods, we improved the generation method of constant variable in algorithm and then put forward an image registration algorithm based on improved GEP, it reduced the randomness of the constant variable in algorithm. The experiment showd that the registration model based on improved GEP has a higher accuracy compared with basic GEP.3. To deal with the imbalanced problems who are a set of problems with highly tilted data, a algorithm based on improved GEP cost-sensitive classification was presented, which threshold value and cost ratio in GEP were taken into consideration and solved the problem of fixed threshold value and cost ratio in the previous algorithm, the experimental result showed that the improved algorithm can obviously improve the classification efficiency of imbalanced problem.4. For the second application use case, the MEP is applied in image retrieval. The image retrieval problem is changed to imbalanced data classification by MEP. After preprocessing image, image contours were extracted, the feature vectors were obtained using image descriptors according to shape features, calculated the distance between the vectors, these distances were used as training data in MEP algorithm, set up MEP model, through which the effective approximated nonlinear similarity function can be designed. It was pointed out by the experiment results that MEP model has a high accuracy in image retrieval. It belongs to a new field of applications for applying MEP in the image retrieval.