Zen Chen

3D shape recovery of complex objects from multiple silhouette images

A reconstruction method is proposed which represents the object with a line-based geometric model. The method does not need the point correspondence information in recovering the 3D object geometry. It is based on the concept of volume intersection, but it is substantially different from the existing octree-based reconstruction methods in the aspects of data structure, reconstruction process and representation uniqueness under a 2D rigid motion. For visualizing the 3D reconstructed object geometry a conversion from the line-based geometric model to a bounded triangular mesh model is developed. The experimental results show that the method is capable of capturing the different details of the object. And it works fast and requires a relatively low memory space.

RECOGNITION OF RADICALS IN HANDWRITTEN CHINESE CHARACTERS BY MEANS OF PROBLEM REDUCTION AND KNOWLEDGE GUIDANCE

Recognizing handwritten Chinese characters is a complex problem. We break the problem down into a series of sub-problems and concentrate on the radical recognition problem. The sub-problems are linked in a hierarchy of three layers: radical, radical sub-unit, and salient stroke. This problem reduction technique enables us to solve complex recognition problems effectively. We describe how to analyze and choose radical sub-units and salient strokes in order to recognize a set of radicals. We also construct two knowledge bases in the form of decision trees to guide the hypothesis and test procedures used to solve the recognition sub-problems. The problem reduction is done in a top-down fashion, while the problem solving process proceeds in a bottom-up fashion. The advantages of our method are described. The representation and organization of the two knowledge bases are explicitly described. The method is tested on two sets of handprinted characters using an IBM PC (486-33). The recognition rate is over 95.7% and the computer time needed to recognize a radical averages about 0.07 s. The experimental results indicate that our method effectively copes with a wide range of ordinary variations encountered in handwritten Chinese characters.

Preclassification of handwritten Chinese characters based on basic stroke substructures

A method for preclassification of handwritten Chinese characters is presented. A set of basic stroke substructures is defined using the consistent stroke connection relations. A knowledge guided recognition process is employed to identify the types of the extracted basic stroke substructures found in a handwritten character. Then a 1-D character coding scheme is given to represent the character and the code can be also used for character preclassification.

A performance controllable octree construction method

The conventional octree construction method is implemented iteratively at consecutive subdivision levels. The resultant octree models at different subdivision levels contain quite different octant compositions, so the system performance, in terms of model accuracy, memory space and construction time, changes widely with the subdivision level number. Since the big system performance gap is not desirable for the practical application, new construction methods must have a finer control over the system performance. In the paper a new construction method is proposed using new types of octree octants. Interesting properties and computer simulations of the new method are presented. The performance comparison between the conventional method and the new method is made under the comparable XOR projection error condition. It is shown that under this condition the memory space and construction time required by the conventional method can be reduced significantly by the new method.

Handwritten Chinese character analysis and preclassification using stroke structural sequence

A unique stroke ordering for handwritten Chinese characters is desirable in many applications including efficient character recognition and automatic radical extraction. Although there are some rules or conventions for writing Chinese characters, yet no consistent and complete rule set is available. Besides special radical knowledge is often needed. It is the purpose of this paper to propose a set of rules for stroke ordering for producing a unique stroke sequence for Chinese characters. It requires no special radical knowledge or knowledge of character block layouts, so it is easy for machine implementation. Moreover, the stroke sequences derived are similar to those given in the dictionary, if not the same. To deal with the writing, variations among writers, we generalize the derived stroke structure sequence to obtain a more consistent stroke information. This generalized stroke structural sequence can be used in the handwritten Chinese character preclassification. Experiments showing applications of our method are reported.

Pose determination of a cylinder using reprojection transformation

Extracting useful and robust feature points from a cylinder image for 3D pose determination is a relatively difficult task, since there are no feature points or lines on the cylinder. In this study, a key vector from a cylinder image is found to yield a transformation. The cylinder is then reprojected from the original image to obtain a new image using the transformation. The new image reveals some important features capable of determining the 3D cylinder position robustly. The proposed method can yield an exact and analytical solution without the requirement of using any special marks, extra equipment or assumptions. Additionally, a problem in which both circles on the cylinder are not completely visible due to the viewing angle can be resolved by this method. Finally, experimental results are provided to verify the theory and show the robustness of our method.

A quality controllable multi-view object reconstruction method for 3D imaging systems

This paper addresses a novel multi-view visual hull mesh reconstruction for 3D imaging with a system quality control capability. There are numerous 3D imaging methods including multi-view stereo algorithms and various visual hull/octree reconstruction methods known as modeling from silhouettes. The octree based reconstruction methods are conceptually simple to implement, while encountering a conflict between model accuracy and memory size. Since the tree depth is discrete, the system performance measures (in terms of accuracy, memory size, and computation time) are generally varying rapidly with the pre-specified tree depth. This jumping system performance is not suitable for practical applications; a desirable 3D reconstruction method must have a finer control over the system performance. The proposed method aims at the visual quality control along with better management of memory size and computation time. Furthermore, dynamic object modeling is made possible by the new method. Also, progressive transmission of the reconstructed model from coarse to fine is provided. The reconstruction accuracy of the 3D model acquired is measured by the exclusive OR (XOR) projection error between the pairs of binary images: the reconstructed silhouettes and the true silhouettes in the multiple views. Interesting properties of the new method and experimental comparisons with other existing methods are reported. The performance comparisons are made under either a comparable silhouette inconsistency or a similar triangle number of the mesh model. It is shown that under either condition the new method requires less memory size and less computation time.

Obtaining base edge correspondence in stereo images via quantitative measures along C-diagonals

Abstract We propose a novel approach to solving the object edge correspondence problem for stereo images. For an object placed on a calibration plate (C-plate), the proposed approach first obtains the correspondences of the object edges lying on the C-plate (the base edges) via quantitative measures of locations of the intersections of the extended line of the edges and the diagonals of the C-plate (C-diagonals) using cross ratios. The measures are viewpoint invariant for an object base edge, and are expressed in the number of image pixels. Special cases which need only coarse calculations, as well as those which require extra measures for additional point features, are also considered. The proposed approach requires 2-D image data only, and is robust in the presence of errors in the feature detection. Experimental results are presented for polyhedral and curved objects to demonstrate the effectiveness of the proposed approach.

A novel 3D planar object reconstruction from multiple uncalibrated images using the plane-induced homographies

Abstract A computer vision method is proposed to determine all the visible 3D planar surfaces in a scene from uncalibrated images and locate them in a single 3D projective space. Most of the existing methods for reconstructing planar objects use point correspondences to estimate the fundamental matrix and derive the compatible projection equations, before they apply the standard triangulation technique to find the 3D points and fit the planes to the 3D points. This type of approaches is generally slow and less accurate because the 3D points are estimated separately, making them vulnerable to image error. We present a plane based reconstruction method to estimate the 3D projective structure using the planar homographies estimated from the plane features in the images. First, we estimate the homography for each visible plane, and then we use the homographies of two primary planes to compute an epipole. We proceed to represent the epipolar geometry for each image pair using the estimated homography and epipole, together with a specified reference plane coefficient vector. Next, we show that the 3D plane coefficient vector of any plane visible in each image pair can be determined with respect to the reference plane coefficient vector once its planar homography is found. Finally, the reconstruction results obtained in individual projective spaces are integrated within a common projective space. To this end, we use the homography and plane equation information of two planes and the epipole associated to derive the coordinate transformation matrix between two involved projective spaces. To evaluate the performance of our method, we apply our method to the synthetic images and real images. All the results indicate the method works successfully.

New Efficient Octree Construction from Multiple Object Silhouettes with Construction Quality Control

In this paper, we propose an octree construction method with a new subdivision strategy which is governed by the degree of overlapping between each generated octant and the object. The specification of a maximum level of subdivision required in the conventional method is not needed. We introduce grey-grey and grey-black nodes based on a construction quality measure. The construction quality is measured by the maximum 2D projection error of the projected octant image with respective to the object silhouette in all views. Only those octants which are grey-grey nodes will be subdivided. Furthermore, we present a fast computation of the 2D projection and a new intersection test to reduce the computer processing time. Computer simulations are conducted to show that the new method outperforms the conventional method in terms of memory space and computation time subject to the same level of construction quality