Research On Adaptive Infrared Image Enhancement Based On Scene Classification

Infrared thermal imager often switches among multiple categories when performing tasks such as tracking and searching.Existing parameter-adjusting adaptive infrared image enhancement algorithms can achieve good enhancement effects for a specific scene.But the adaptive enhancement effects for different locations is limited.Experiments have shown that the infrared radiation characteristics of forests,deserts,oceans,cities and other different scenes are significant different.To meet the various image enhancement wishes,it is of great practical value to study an adaptive infrared image enhancement algorithm on scene classification.The collected infrared image dataset in the thesis is divided into three typical scenes according to the actual application scenarios: aerial scene,land scene,and sea scene.Based on the study of the infrared radiation characteristics of different scenes,three different infrared image enhancement algorithms are proposed.According to the image classification results,the corresponding algorithm is automatically matched to achieve adaptive enhancement of infrared images in different scenes.The main research contents of this thesis are as follows:Regarding scene classification,the number of peaks,peak width,and kurtosis in the gray-level histogram curve of the infrared image are statistically extracted.And a method using threshold judgment to distinguish scenes is proposed.Then,the features such as Gray-Level Co-occurrence Matrix,Local Contrast Histogram,Histogram of Oriented Gradient,and Local Binary Pattern Histogram in image classification are studied.Finally,through experiments,the polynomial kernel SVM method based on Histogram of Oriented Gradient is determined to having the highest scene classification accuracy.Regarding image enhancement algorithms,an improved non-uniformity correction method is used to effectively reduce the "veiling effect" on human visual perception.Secondly,a background suppression algorithm based on guided filtering is used for aerial scene infrared images.A detail enhancement algorithm based on guided filtering hierarchy is used for land scene infrared images.An improved segmented mapping algorithm based on generalized histogram equalization is used for sea scene infrared images.To verify the feasibility of the system,a software for a multi-scene adaptive infrared image enhancement system is developed using the MATLAB App Designer.System testing indicates that the software can achieve adaptive enhancement of infrared images in multiple scenes.