Research On Estimation Illumination Direction Using Multiple Features

Illumination intensity,color and direction are the mainly parts in illumination conditions.If illumination direction is known in the scene,it can avoid the shadow interference in the object tracking or recognition process,and produce a more realistic effect in the augmented reality.Estimating illumination direction has been an important research topic in computer vision.There are already a lot of literature about the illumination direction,but these methods have some limitations.(1)The reference objects whose shape are known need to be placed in the scene;(2)Some methods need to manually tag the feature points of object and the corresponding points of the shadow in the scene;(3)Some methods need to know the three-dimensional structure of objects.A new method is presented to estimate illumination direction in combination with multiple features,no requiring a specular sphere or given geometric structure of real objects.The main work of this thesis includes three parts:A new method is presented to estimate illumination direction on the basis of the differential principle,no requiring a specular sphere or given geometric structure of real objects.According to the principle of differential and the corresponding relationship between the center of gravity,the shadow and shading regions are fragmented into corresponding rectangular areas,forming pairs of shadow-shading patches.Projection model are built with the pairs of patches to estimate azimuth and zenith angles of illumination direction.The average of these azimuth and zenith angles is considered as the illumination direction of the scene.Experimental results on the datasets of indoor and outdoor scenes demonstrated that our method is efficient to estimate illumination direction of a scene,without geometric information of the objects.Estimating illumination direction needs to detect the shadow,a method is proposed to separate object and shadow in combination with intensity,chrominance and texture of the objects.This method is the improvement of the adaptive shadow estimator method.Our method joins the texture information into the adaptive shadow estimator method,and combines intensity,chrominance and texture information by means of weighted.Experimental results show that the algorithm can effectively separate object and shadow.Estimating illumination direction needs to get the shading region.For the problem that the shading region is difficult to detect,we estimate the shading regionson the basis of dark channel priori,intensity,texture and color.The dark channel describes the intensity information.The intensity of shading region is lower than the illumination region,and the threshold value is obtained between the shading and illumination region using the Otsu threshold method.We get the initial shading region by the threshold value,and then detect the shading region in combination with intensity,texture and color information.We make the assumption that if the regions have the same texture and chrominance information,then they have the same material properties.The region that has higher brightness value is considered as the illumination region,otherwise as the shading region.Experimental results on the datasets of indoor and outdoor scenes show that the method can effectively separate the shading region.