Research On Enhancement Of Degrade Images In Foggy Traffic

Fog is a weather phenomenon.Even on a sunnyday,we can see that an object at a distance is affected by fog.Fog,which is caused by water vapor,is also a problematic weather,and is listed in top ten problematic weathers.It causes many bad results to traffic,aviation,traveling,our daily lives and our health.For the military,foggy images might bring distortion that can result in serious event.Traffic is affected by weather in a most serious way.The driver's view is wreaked,and in many cases it will result in a traffic accident.So how to improve a driver's view in foggy weather has become the focuse of academics in information and ITS science.Many scientist and engineers have studied it a lot,and many great achievements have been made so far.But the problem is very difficult and complicated and the information about objects in foggy traffic is insufficient.Currently the degraded process of a foggy image can not be described by any algorithms or models perfectly. There is still much works to do on this research subject,so it has a great significance to study how to enhance the foggy traffic images.In this thesis,we analyzed the features of objects in foggy traffic and the algorithm and models which are used to enhance the degraded images.The main works completed are as follows:(1) Analyzed the color and spectrum features of objects in foggy traffic, extracted their characteristics by histogram,spastics methods.With the analyzing of spectrum,energy,abstract dimensions,we got many useful traits or characteristics, such as center,horizontal,and vertical low frequency energy,multi-scale sub-space energy distribution,and fractal dimensions.(2) Proposed one restoration algorithm for foggy traffic images by the combination of the Retinex theory and the sky area.The sky area was detached from the traffic images by EM or FCM.In one case,the PCA was used to get a nearly perfect result.After detaching the sky area successfully,the illumination was reconstructed by the information of the sky areas.(3) Proposed two methods which can be used to estimate the parameters in the atmospheric scattering model.The first one is the improved algorithm using one foggy image.First,we get the rough estimations of the object's images by simple methods.Then by substituting them into the atmospheric scattering equation,a more effective solution was obtained this time.By repeating this process,a satisfying solution will be obtained.The second one is the improved algorithm with multi-images(usually two).The two images were taken in the same place under different weather condition.The contours were discrete and the problem becomes a classify problem.The problem can be solved by defined distance inner class and other features with FCM.(4) Proposed two new models based on the atmospheric scattering model and total variation model.The first one was constructed by adding the atmospheric scattering model into the total variation minimization models as a restricting condition.The second one was constructed by using the atmospheric scattering model as the comparability condition,combining it with the total variation and contrast enlargement.(5) Proposed several fast enhancement algorithms for foggy traffic images.The first one provides improved histogram equalization.The main idea for this algorithm is that the contrast is decreased in sky area,and the contrast enlargement is proportional to the distance from a point in the objects to camera.The second one is the fast enhancement algorithm with depth of field compensation in objects.While keeping the advantage in dealing with local high frequency data for histogram equalization,the scattering as depth varying was considered.The third one is the fast algorithm based on the simplified SSR method.In SSR,a lot of convoluted computations are needed,and this processing needs a long computing time.In this thesis,we will present a new,simplified,SSR algorithm that will markedly decrease the computing task.