| Summary: | The development of a satisfactory the three-dimensional image system is a constant pursuit of the scientific community and entertainment industry. Among the many different methods of producing three-dimensional images, integral imaging is a technique that is capable of creating and encoding a true volume spatial optical model of the object scene in the form of a planar intensity distribution by using unique optical components. The generation of depth maps from three-dimensional integral images is of major importance for modern electronic display systems to enable content-based interactive manipulation and content-based image coding. The aim of this work is to address the particular issue of analyzing integral images in order to extract depth information from the planar recorded integral image. To develop a way of extracting depth information from the integral image, the unique characteristics of the three-dimensional integral image data have been analyzed and the high correlation existing between the pixels at one microlens pitch distance interval has been discovered. A new method of extracting depth information from viewpoint image extraction is developed. The viewpoint image is formed by sampling pixels at the same local position under different micro-lenses. Each viewpoint image is a two-dimensional parallel projection of the three-dimensional scene. Through geometrically analyzing the integral recording process, a depth equation is derived which describes the mathematic relationship between object depth and the corresponding viewpoint images displacement. With the depth equation, depth estimation is then converted to the task of disparity analysis. A correlation-based block matching approach is chosen to find the disparity among viewpoint images. To improve the performance of the depth estimation from the extracted viewpoint images, a modified multi-baseline algorithm is developed, followed by a neighborhood constraint and relaxation technique to improve the disparity analysis. To deal with the homogenous region and object border where the correct depth estimation is almost impossible from disparity analysis, two techniques, viz. Feature Block Pre-selection and “Consistency Post-screening, are further used. The final depth maps generated from the available integral image data have achieved very good visual effects.
|
|---|
