Research Onalgorithm Of SFS3D Reconstruction Based On Self-shadow Processing

With the continous development of computer vision and virtual reality,3Dreconstruction of images is becoming an important study day by day, and plays moreand more important status and role in the various fields of production and life insociety, especially in industrial, medical, national defence, archeology, etc, havingvery wide application value. At present, many reconstruction technologies have beenformed, and are generally designated “Shape from X” technologies. X can be motion,texture, contour, shadow, etc. One of the technologies, Shape from Shading (SFS forshort), only need gray level information of one image to recover3D shape of objects.This technology is relatively easy to achieve, only needs simple hardware devices andexternal conditions, and can be used in some special occasions in which otherthechnologies is not suitable.Traditional SFS algorithms most adopt harsh assumptions that don’t accord withactual situation, some disturbed greatly by noise, some with very complex calculation,some with slow convergence speed and low efficiency, and some with small suitablescope. The main work and innovation of this paper is to improve the detection methodof the self-shadow, and optimize cumulation algorithm of least element to achieve3Dreconstruction of objects.For the problem of self-shadow, this paper mainly studys a method of the edgedetection based on the model. This method approximately illuminant direction, andthen detects self-shadow area by the edge detection and amends. It removes actualself-shadow area, and effectively solves a phenomenon that traditional methodseliminate some information of objects when the color of self-shadow is similar to thecolor of objects. So it avoids producing big distortion to the result of3Dreconstruction.Generally cumulation algorithm of least element does not consider theself-shadow of objects, and assume that Albedo value is a constant and light sourcedirection is known, while this paper optimizes it. This paper chooses the curvedsurface which conforms to the assumptions to estimate the light source direction,adopts local estimating method to estimate albedo value after the image is segmentalized, adopts nonlinear least square method to solve cumulation algorithm ofleast element, and uses the method of two recursive path computing and weightedaverage to enhance the accuracy of reconstruction. So, this paper obtains betterreconstruction results.At last, this paper compiles corresponding3D reconstruction programs byMatlab software, respectively reconstructs synthesis images and actual images. Theresults of the experiments verify that the algorithm in this paper is more precise andstable than generally cumulation algorithm of least element and traditional algorithms.The algorithm in this paper is wide application, and can be applied to the objects withself-shadow and more complicated shape.