On the basis of a previous encoding pattern with a large size M-array, this paper researches how to decode a captured image and then to implement the image matching and image reconstruction. The rectangular block, circle and the disc are chosen as M-array symbols that are easily indentified and distinguished to encode the projecting pattern, then , they are used to study the decoding method for the captured encoding M-array image. By considering projection patterns, symbol features, environments, symbol distortion and other factors fully, the decoding scheme is determined. After preprocessing the captured image, the symbol feature recognition is adopted to recognize the classification of symbols, calculate the directions of angular blocks, and to obtain the directions of circle and disc and determine the nether neighborhood and the right neighbor of symbols by an adjacent angular block as the reference. Furthermore, the decoded M-array is gotten by traversing all symbols. According to the window uniqueness, the position of each window in the projection pattern is determined, then the location of each match point is found and a coordinate matrix of matched points is generated. Finally, the depth information of 3D surface is obtained by using triangulation method and the reconstruction of a 3D surface is finished.