Research On Low-light Image Colorization Method Based On Multi-dimensional Data Association Rules Mining

At present,most of the imaging effects of military reconnaissance device,such as low light level night vision device and infrared imager,and some safety inspection equipments are gray scale images.The ability to distinguish between colors of human eyes is higher than the ability to distinguish between gray levels,and monochromatic image colorization can greatly improve the ability of detection and recognition of the targets.So the research of image colorization has a wide application prospect in both military and civil fields.In this paper,we have studied the related theories of association rules and related mining algorithms.On this basis,we have focused on the application of association rule mining in the field of low light level image colorization.The main contributions of this article are as follows:(1)This article applies association rule mining in the field of microchromic image colorization for the first time,and proposes a new method of low light level image colorization based on combined association rule set mining.Firstly,it uses the idea of"binning" for the image brightness and color components to shorten the time cost of Apriori algorithm,and generates a combination of "brightness-category-color" association rule set from the reference color image;Then,the SVM classifier is trained to classify the pixel in the target image;Lastly,searching and mapping are carried out in the combined association rule set to generate the result.Experiments show that the method proposed in this paper has a good colorization performance..(2)This paper also proposes a method of low light level image colorization by based on the classification association rule set mining.Firstly,it selects the corresponding reference color image for the objects existing in the target image and generates a single association rule set corresponding to different objects;Then,The pixels in the target image are classified libSVM classification algorithm,and searching and mapping are carried out in the corresponding association rule set by the brightness of pixels;Finally,colorization result can be generated.It also puts forward the idea of "off-line association rule library",which can be uploaded every time when a association rule set corresponding to an object is excavated,and the association rule set can be directly invoked later.Experiments show that the method can achieve a very fast and good-effect colorization process.This method proposed in this paper has high practical value,which has high speed,good effect,and the association rules can be reused.