3D shapes of objects or scenes in an image can be obtained from its image or images. The shape-from-shading (SFS) technique uses the pattern of shading in the image to obtain shape information. In this thesis, a hybrid SFS method based on methods which are included in the linearization-based classification of SFS algorithms is presented. Our aim is to use the Fouirer coefficients of central differences obtained from the gray-level image as prior knowledge. By using functionally generated surfaces and different data sets in the literature, we compare the proposed method with other linearization-based approaches. Five different evaluation metrics were applied to recovered depth maps and corresponding gray-level images. We tabularly and graphically present the results, and also specify required time. As a result of the tests, the proposed method can prevent the fluctuations on the surfaces and obtain better 3D reconstruction results at low parameters.
In the second stage of this thesis, the developed SFS method is specialized for the quality control of metallic parts, which is a real world industrial problem. An image acquisition system is proposed to capture the images of metal components and shape information are obtained by using SFS methods. Then, with the help of pre-processing steps and morphological methods, defective areas of the surface are detected by using the estimated shape information. Real-time monitoring of the results is also provided with a developed interface.
Finally, a solution to the ambiguity of the surfaces being concave or convex, which is one of the main problems of SFS methods, is proposed in 3D modeling environment. Color patterns, formed on the surface illuminated with different colors of illumination from different directions, are processed algorithmically and information about the orientations of the surface regions are obtained. We show in detail that five different SFS methods make more accurate surface estimation with the help of developed algorithm. |