| The road information management has be applied to the actual situation gradually as the country building roads energetically.Camera is the major equipment for data acquisition in the process of road information management.Only by ensuring that camera works normally can the road information management runs one the rail.This thesis is researching images of color cast in the road video for detecting and correcting.First,this thesis proposes a partition-based color cast detection model(PBCCD).The image of road video has some characteristics,such as neutral hue,uneven hue distribution,low contrast,low saturation and indeterminate color cast type by analyzing the background of this thesis in theory of colours.The traditional algorithms can detect general images color cast rightly in road video,but they can’t detect localized images color cast rightly.So,a color cast detection model is proposed in this thesis.This model takes the YUV color space appropriately,and applies gray theory in YUV to separate the road background from the prospect.Then,PBCCD splits images by DFS algorithm for dividing and conquering,to process localized images color cast effectively.The PBCCD model uses SVM method to classify color images and normal images in two categories by eigenvalues of images in YUV color space,because YUV color space can reflect the regular color cast of road videos more effectively.Secondly,an Adaptive Gain Multi-Scale Retinex with Color Restoration(AG-MSRCR)algorithm is proposed.After the experimental comparison,MSRCR algorithm has a good correction effect,but will retain the image color cast data.So,this thesis proposes AG-MSRCR algorithm.The fixed color gain parametersβin the original algorithm are changed to adjust the adaptive gainβ_i under log-domain data so that they can gain color according to the overall situation of the image to solve the image color imbalance caused by color cast problem.Through experimental verification,the correcting color cast images of the proposed AG-MSRCR algorithm has better effect.Combine the AG-MSRCR algorithm with the PBCCD model to correct color of the segmented image and merge the restored images.This thesis proposes an image linear splicing method that linear splices images based on an over-region,making the resulting image smoother in the spliced part to eliminate the obvious boundary problem of stitched images.Finally,the PBCCD model and the AG-MSRCR algorithm are implemented in OpenCV environment and are analyzed their performance.The PBCCD model is compared with existing color cast detection algorithms such as GW and equivalent circle.The accuracy of color cast detecting in all images is 94.99%by the PBCCD model,which is higher than the existing color cast detection algorithm.The AG-MSRCR algorithm has a better correction effect than the existing algorithms such as GW correction and MSRCR. |
