Research And Application Of Image Fusion Algorithm Based On Structured Dictionary Learning

As time goes,the way people get information is gradually transformed into image acquisition.In order to meet people's needs for the image,we developed a variety of scenes,different functions of the imaging sensor.The imaging mechanism of each imaging sensor is different,and the image content is different.In order to make the quality of the image can meet the various needs,image fusion as a comprehensive use of the characteristics of the image,the sensor imaging of the complementary information into a more rich information can be an important technical means,is widely used in various Areas such as military,medical disease diagnosis,remote sensing and other fields.Based on the sparse representation theory,this paper proposes a method of multi-sensor image fusion.The main findings are as follows:Firstly,according to the characteristic of rich structure information in natural images,an adaptive learning structured dictionary learning algorithm is proposed in the sparse representation dictionary learning algorithm.Sparse representation of the image coding theory as a similar visual cortical neuron.Sparse representation with over-complete dictionary as a base to represent the image only need a small amount of dictionary atoms can reveal the essence of image content.In this paper,by studying the characteristics of the dictionary in the representation of the image signal,that is,the structure of the image information needs to combine some of the atoms to its representation.According to this feature,the group information of the dictionary can be adaptively obtained in the course of training through the rule of sparse representation coefficients.In this paper,we study the properties of sparse representation coefficients,and make the structured dictionary information more accurate,so that the dictionary representation capability is greatly improved.And then analyzes the group structure in the dictionary learning algorithm which is obtained by the similarity of the dictionary atoms.Such a grouping process will lead to higher atomic similarity in the dictionary,especially the correlation of atoms in the group will be high.In order to overcome the above problems in the process of learning to add a relevance of the judge to avoid this situation.That is,when the group's atomic similarity is too high to remove the atoms and then add new atoms to supplement.Through this technique,the diversity of the learned dictionaries can be improved,so that the presentation ability of the dictionary can be improved remarkably.Finally,we analyze the existing problems of fusion rules based on sparse representation of image fusion,that is,the image obtained by fusion of L1 norm maximum fusion rules will lose some significant information in the fused image.By analyzing the physical meaning behind the existing structured dictionaries,each group of dictionaries represents the structural information in the image content.So in the integration can be deep into the structure of the image to determine the characteristics of the fusion.According to the above analysis,the fusion rule based on structural the maximum L1 norm is proposed.The fusion rules get the fusion image as much as possible to preserve the significant information in each source image.