Research On Day-or-night Image Classification And Image Defogging

Optical imaging of outdoor scene changes effected by fog,especially by week light condition such as night,which lead to low brightness,low contrast,color distortion,et al and which severely restricted and affected the play of target detection,recognition and other visual systems.On the other hand,the imaging mechanism vary greatly influenced by light and temperature and which brought great difficulties to visual systems which require continuous work.So it is very important to study accurate image recognition algorithm and nighttime image defog algorithm.The imaging mechanism is different between daytime image and nighttime image.Distinguishing methods based on human's eyes can not meet the requirements of automation,continuous work and robustness;the methods based on time,characteristics of objects characteristics or statistical law of fixed scene have poor adaptability and can not satisfy different scenes.Either,hardware method have high cost and poor accuracy robustness.Nighttime degrade Most exiting defog algorithm is designed for daytime image and there are less research on nighttime image defog.In view of the above situation,a recognition method based on statistical features was proposed.Through analysis statistical features from a large number images and fuse multiple features which can highlight the difference between daytime image and nighttime image,we get recognition method which can achieve accurate result.In the other hand,nighttime image defog methods have high degree of distortion,blurring and poor stability,this paper proposed two nighttime fog imaging models and defog algorithm correspondingly According to the high similarity between inverse image of low illumination image and daytime fog image,nighttime fog image can be approximated as the reverse image which is operated by reverse and add fog sequentially to the daytime fog image,that is the model I,thus the nighttime image defog algorithm was proposed based on dark channel prior and color correction.Alse basing on the atmospheric scattering model and the theory that clear nighttime image can be divided into the image generated by artificial light and the image generated by natural light,which is model II,and then the defog algorithm was proposed based on the dark channel prior and multi-scale retinex.