3D shape reconstruction using 2D images

Many methods are used to obtain three-dimensional shapes from two-dimensional images. Shape from shading and photometric stereo methods use shadow information in images to obtain three-dimensional shape information. Each method has its own strengths and weaknesses. Shape from shading method gives better results for detailed regions such as mouth, eyes and nose. Photometric stereo method gives better results for the general shape information. To obtain better depth information these two methods are combined in some studies. In this study, applied methods are investigated and a better combination method is presented. And also during this study two new fusion methods are explained and test results are presented.Firstly, depth information is obtained from the shape from shading and photometric stereo. The calculated depth information is used as input for the application.In threshold value method for combining, the depth information obtained from the shape from shading and photometric stereo is compared in each point. For each point value, if photometric stereo value greater than shape from shading value, the result depth value is photometric stereo?s depth value. Otherwise, a pre-defined limit value is determined and is used as criteria for comparison. If shape from shading depth value is greater than photometric stereo depth value and also their difference is smaller than the threshold value, the result value is shape from shading?s depth value, if the value of difference is greater than the threshold, 30% of the depth of shape from shading and 70% of the depth of photometric stereo are summed together for the result of depth value.In the other combination method, the depth maps are separately normalized. Then for every point, maximum value is added and the logarithm of each depth information is separately taken. Result of the depth information is obtained from the shape from shading and photometric stereo depth information with certain ratios. These rates are 20% for shape from shading and 80% for photometric stereo.Errors of depth information are calculated by using the mean square and the mean average error methods. When these errors are investigated, it is seen that the proposed approaches are successful.