Сегментация исходных снимков для фотометрического метода трехмерного сканирования

Abstract

Purpose: Previous to surface gradient calculation by photometric stereo, it is necessary to prevent any inconsistency between the used reflectance model and the input data. The most effective way free from any object pre-processing is the fragmentation of the source images. The purpose of the work is to define and systematize the features of shadows, highlights and reflections, and to develop a segmentation algorithm which would take into account the specificity of photometric stereo. Methods: The experimental data were obtained by computer simulation of source images using the Phong reflection model. For image segmentation, we used the features of highlights and shading based on verified linear dependence of the intensities of images from the same plane and on the minimized function of the deviation between the input data and the model-fitting data. Results: A photometric 3D scanner configuration is defined, consisting of five fixed light sources. With this configuration, you can use additional features of shadows, interreflections and highlights during the image segmentation. The redundant number of images ensures that at least three intensities at each visible point are well verified. An algorithm was developed for checking the conformity between the input data and the light reflection model used for the calculation. In this algorithm, for the majority of the visible points, the problem of two-variable function minimization needs no solution or can be reduced to the problem of single variable function minimization. A result of the segmentation is shown for two objects. One of these objects is modeled with highlights and surface gaps, which often make it difficult to calculate the surface orientation. Practical relevance: Compared to its analogues, the developed algorithm reduces the time required for the visible area segmentation and provides more elaborate segmentation of the source images.