Tian-Li Yu

Kernelized structural SVM learning for supervised object segmentation

Object segmentation needs to be driven by top-down knowledge to produce semantically meaningful results. In this paper, we propose a supervised segmentation approach that tightly integrates object-level top down information with low-level image cues. The information from the two levels is fused under a kernelized structural SVM learning framework. We defined a novel nonlinear kernel for comparing two image-segmentation masks. This kernel combines four different kernels: the object similarity kernel, the object shape kernel, the per-image color distribution kernel, and the global color distribution kernel. Our experiments show that the structured SVM algorithm finds bad segmentations of the training examples given the current scoring function and punishes these bad segmentations to lower scores than the example (good) segmentations. The result is a segmentation algorithm that not only knows what good segmentations are, but also learns potential segmentation mistakes and tries to avoid them. Our proposed approach can obtain comparable performance to other state-of-the-art top-down driven segmentation approaches yet is flexible enough to be applied to widely different domains.

A Genetic Algorithm for Developing Modular Product Architectures

The architecture of a product is determined by both the elements that compose the product and the way in which they interact with each other. In this paper, we use the design structure matrix (DSM) as a tool to capture this architecture. Designing modular products can result in many benefits to both consumers and manufacturers. The development of modular products requires the identification of highly interactive groups of elements and arranging (i.e. clustering) them into modules. However, no rigorous DSM clustering technique can be found in product development literature. This paper presets a review of the basic DSM building blocks used in the identification of product modules. The DSM representation and building blocks are used to develop a new DSM clustering tool based on a genetic algorithm (GA) and the minimum description length (MDL) principle. The new tool is capable of partitioning the product architecture into an “optimal” set of modules or sub-systems. We demonstrate this new clustering method using an example of a complex product architecture for an industrial gas turbine.Copyright

Genetic algorithm design inspired by organizational theory: pilot study of a dependency structure matrix driven genetic algorithm

This study proposes a dependency structure matrix driven genetic algorithm (DSMDGA) which utilizes the dependency structure matrix (DSM) clustering to extract building block (BB) information and use the information to accomplish BB-wise crossover. Three cases: tight, loose, and random linkage, are tested on both a DSMDGA and a simple genetic algorithm (SGA). Experiments showed that the DSMDGA is able to correctly identify BBs and outperforms a SGA.

Dependency structure matrix, genetic algorithms, and effective recombination

In many different fields, researchers are often confronted by problems arising from complex systems. Simple heuristics or even enumeration works quite well on small and easy problems; however, to efficiently solve large and difficult problems, proper decomposition is the key. In this paper, investigating and analyzing interactions between components of complex systems shed some light on problem decomposition. By recognizing three bare-bones interactionsmodularity, hierarchy, and overlap, facet-wise models are developed to dissect and inspect problem decomposition in the context of genetic algorithms. The proposed genetic algorithm design utilizes a matrix representation of an interaction graph to analyze and explicitly decompose the problem. The results from this paper should benefit research both technically and scientifically. Technically, this paper develops an automated dependency structure matrix clustering technique and utilizes it to design a model-building genetic algorithm that learns and delivers the problem structure. Scientifically, the explicit interaction model describes the problem structure very well and helps researchers gain important insights through the explicitness of the procedure.

Population sizing for entropy-based model building in discrete estimation of distribution algorithms

This paper proposes a population-sizing model for entropy-based model building in discrete estimation of distribution algorithms. Specifically, the population size required for building an accurate model is investigated. The effect of selection pressure on population sizing is also preliminarily incorporated. The proposed model indicates that the population size required for building an accurate model scales as Θ(m log m), where m is the number of substructures of the given problem and is proportional to the problem size. Experiments are conducted to verify the derivations, and the results agree with the proposed model.

Efficient Message Representations for Belief Propagation

Belief propagation (BP) has been successfully used to approximate the solutions of various Markov random field (MRF) formulated energy minimization problems. However, large MRFs require a significant amount of memory to store the intermediate belief messages. We observe that these messages have redundant information due to the imposed smoothness prior. In this paper, we study the feasibility of applying compression techniques to the messages in the min-sum/max-product BP algorithm with 1D labels to improve the memory efficiency and reduce the read/write bandwidth. We articulate properties that an efficient message representation should satisfy. We investigate two common compression schemes, predictive coding and linear transform coding (PCA), and then propose a novel envelope point transform (EPT) method. Predictive coding is efficient and supports linear operations directly in the compressed domain, but it is only compatible with the L1 smoothness function. PCA has the disadvantage that it does not guarantee the preservation of the minimal label. EPT is not limited to L1 smoothness cost and allows a flexible quality vs. compression ratio tradeoff compared with predictive coding. Experiments on dense stereo reconstruction have shown that the predictive scheme and EPT can achieve 8times or more compression without significant loss of depth accuracy.

Using human body gestures as inputs for gaming via depth analysis

Natural ways of input greatly enhance the entertainment experience for emerging gaming systems represented by the Sony Playstation 2 EyeToy and the Nintendo Wii Console. In this paper we present a novel method of using human body gestures depth image as gaming application input. Depth images have natural advantages over grayscale or color images in terms of robustness against illumination change, texture complexity, and background interference. Our proposed method consists of three major components: depth image acquisition, mean shift based preprocessing, and HMM-based gesture recognition. We validate our method by applying it to a boxing game scenario to distinguish boxing gestures such as dodge, jab, hook, and uppercut. The experiment results indicate that our method can efficiently distinguish the subtle differences among these gestures and yield excellent accuracy (up to about 98%). The potential usage of the proposed method on gaming applications and generic human computer interaction is very promising.

Conquering hierarchical difficulty by explicit chunking: substructural chromosome compression

This paper proposes a chromosome compression scheme which represents subsolutions by the most expressive schemata. The proposed chromosome compression scheme is combined with the dependency structure matrix genetic algorithm and the restricted tournament replacement to create a scalable optimization tool which optimizes problems via hierarchical decomposition. One important feature of the proposed method is that at the end of the run, the problem structure obtained from the proposed method is comprehensible to human researchers and is reusable for larger-scale problems. The empirical result shows that the proposed method scales sub-quadratically with the problem size on hierarchical problems and is able to capture the problem structures accurately.

Recovering shape and reflectance model of non-lambertian objects from multiple views

This paper proposes an algorithm to simultaneously estimate both the 3D shape and parameters of a surface reflectance model from multiple views of an object made of a single material. The algorithm is based on a multiple view shape from shading method. A triangular mesh represents the shape of the object. The Phong reflectance model is used to model the surface reflectance. We iteratively find the shape and reflectance parameters that best fit all input images. Subdividing triangles in the mesh into smaller ones gradually refines the estimates of shape and reflectance model. The estimation takes into account both self-occlusion and self-shadowing. Analysis shows that the accuracy of reflectance estimation is limited by the triangle size in the shape model. We also propose to use Richardson extrapolation to overcome this and further refine the reflectance model estimate. The estimated 3D shape and reflectance model can be used to render the same object from different viewing directions and under different lighting conditions. Experimental results on both synthetic and real objects are given.

Linkage learning, overlapping building blocks, and systematic strategy for scalable recombination

This paper aims at an important, but poorly studied area in genetic algorithm (GA) field: How to design the crossover operator for problems with overlapping building blocks (BBs). To investigate this issue systematically, the relationship between an inaccurate linkage model and the convergence time of GA is studied. Specifically, the effect of the error of so-called false linkage is analogized to a lower exchange probability of uniform crossover. The derived qualitative convergence-time model is used to develop a scalable recombination strategy for problems with overlapping BBs. A set of problems with circularly overlapping BBs exemplify the recombination strategy.