DEVELOPMENT OF A METHOD FOR DIGITAL RECTIFICATION OF IMAGES

Abstract

Most historical buildings have been destroyed or damaged by natural phenomena and adverse human activities. Their reconstruction pose a great problem owing to the lack of architectural records in most cases. Historical photographs are the only means available for their reconstruction. These photographs neither have the camera information nor any reference points. A plane object (or one whose relief does not exceed a certain tolerance) can be processed without using a stereopair. Photogrammetric surveys based on one image are derived through some kind of rectification. Historically, rectification has been carried out through graphic or optical-photographic techniques. Today, these techniques have been replaced by analytical, and mostly digital rectification. Photogrammetric single image techniques, like the generation of rectified images are well suited for use in architecture as well as for monument preservation. They combine true scale geometric measurements with full image information under inexpensive production costs. Especially in the field of architectural photogrammetry, the combination of image processing with photogrammetric systems provides new solutions. There are different approaches for the photogrammetric rectification of images. The simplest method is the projective rectification, extensively usedjn the^ past with analogue rectifiers. With digital image processing, some new rectification techniques have become possible, like polynomial rectification or unwrapping provide new interesting and non-expensive solutions in architectural photogrammetry. Analytical methods of rectification also have many limitations. To overcome the limitation, a methodology was developed for digital rectification of images in the present research work, based on finite element method. To validate the methodology it was tested for one vanishing point perspective and two vanishing point perspective. After testing the effectiveness of the methodology for these two cases, these were compared with the projective transformation, affine transformation and polynomial of order 2. The comparison was done in different phases following the elimination rule. Finally the proposed method was tested with projective transformation first for a practical case and than for a fictitious image generated using perspective geometry to compare the absolute accuracy of the two methods. The proposed method provides accurate rectified image than the normally used analytical methods for rectification e.g. projective transformation, affine transformation and polynomial of order two. Generally, most analytical methods work well in interpolation only. The proposed method requires only four control points, same as required in the case of projective transformation. It whereas divides the entire image into number of pieces on the basis of perspective geometry, extending the control over the whole image, and rectifies the entire image piece by piece. Hence, it also controls the errors in extrapolation. The method is tested on images of tall building, sculpture, and building with pillars and generated grid mesh. The maximum shift observed on the farthest grid point, which is about 5.5 times the diagonal grid distance, is limited to 2.5 pixels and the rmse is ±1.8 pixels. VI maximum shift observed on the farthest grid point, which is about 5.5 times the diagonal grid distance, is limited to 2.5 pixels and the rmse is ±1.8 pixels. Since, finite element method uses inverse method of rectification, the resampling of rectified image is also not required by the proposed method. The errors occurred due to resampling by other analytical methods may also be avoided by this method. The proposed method has a greater application in mapping of buildings and monuments of architectural importance using minimum number of control points on a single photograph. Thus, it may save a considerable amount of time and money. The method could also be used for the registration of thematic maps and satellite images. Vll