Jia-Guang Sun

Transfer Feature Learning with Joint Distribution Adaptation

Transfer learning is established as an effective technology in computer vision for leveraging rich labeled data in the source domain to build an accurate classifier for the target domain. However, most prior methods have not simultaneously reduced the difference in both the marginal distribution and conditional distribution between domains. In this paper, we put forward a novel transfer learning approach, referred to as Joint Distribution Adaptation (JDA). Specifically, JDA aims to jointly adapt both the marginal distribution and conditional distribution in a principled dimensionality reduction procedure, and construct new feature representation that is effective and robust for substantial distribution difference. Extensive experiments verify that JDA can significantly outperform several state-of-the-art methods on four types of cross-domain image classification problems.

Control Point Adjustment for B-Spline Curve Approximation

Pottmann et al. propose an iterative optimization scheme for approximating a target curve with a B-spline curve based on square distance minimization, or SDM. The main advantage of SDM is that it does not need a parameterization of data points on the target curve. Starting with an initial B-spline curve, this scheme makes an active B-spline curve converge faster towards the target curve and produces a better approximating B-spline curve than existing methods relying on data point parameterization. However, SDM is sensitive to the initial B-spline curve due to its local nature of optimization. To address this, we integrate SDM with procedures for automatically adjusting both the number and locations of the control points of the active spline curve. This leads to a method that is more robust and applicable than SDM used alone. Furthermore, it is observed that the most time consuming part of SDM is the repeated computation of the foot-point on the target curve of a sample point on the active B-spline curve. In our implementation, we speed up the foot-point computation by pre-computing the distance field of the target curve using the Fast Marching Method. Experimental examples are presented to demonstrate the effectiveness of our method. Problems for further research are discussed. q 2003 Elsevier Ltd. All rights reserved.

Transfer Joint Matching for Unsupervised Domain Adaptation

Visual domain adaptation, which learns an accurate classifier for a new domain using labeled images from an old domain, has shown promising value in computer vision yet still been a challenging problem. Most prior works have explored two learning strategies independently for domain adaptation: feature matching and instance reweighting. In this paper, we show that both strategies are important and inevitable when the domain difference is substantially large. We therefore put forward a novel Transfer Joint Matching (TJM) approach to model them in a unified optimization problem. Specifically, TJM aims to reduce the domain difference by jointly matching the features and reweighting the instances across domains in a principled dimensionality reduction procedure, and construct new feature representation that is invariant to both the distribution difference and the irrelevant instances. Comprehensive experimental results verify that TJM can significantly outperform competitive methods for cross-domain image recognition problems.

Detecting implicit dependencies between tasks from event logs

Process mining aims at extracting information from event logs to capture the business process as it is being executed. In spite of many researchers’ persistent efforts, there are still some challenging problems to be solved. In this paper, we focus on mining non-free-choice constructs, where the process models are represented in Petri nets. In fact, there are totally two kinds of causal dependencies between tasks, i.e., explicit and implicit ones. Implicit dependency is very hard to mine by current mining approaches. Thus we propose three theorems to detect implicit dependency between tasks and give their proofs. The experimental results show that our approach is powerful enough to mine process models with non-free-choice constructs.

Transfer Sparse Coding for Robust Image Representation

Sparse coding learns a set of basis functions such that each input signal can be well approximated by a linear combination of just a few of the bases. It has attracted increasing interest due to its state-of-the-art performance in BoW based image representation. However, when labeled and unlabeled images are sampled from different distributions, they may be quantized into different visual words of the codebook and encoded with different representations, which may severely degrade classification performance. In this paper, we propose a Transfer Sparse Coding (TSC) approach to construct robust sparse representations for classifying cross-distribution images accurately. Specifically, we aim to minimize the distribution divergence between the labeled and unlabeled images, and incorporate this criterion into the objective function of sparse coding to make the new representations robust to the distribution difference. Experiments show that TSC can significantly outperform state-of-the-art methods on three types of computer vision datasets.

A workflow net similarity measure based on transition adjacency relations

Many activities in business process management, such as process retrieval, process mining, and process integration, need to determine the similarity or the distance between two processes. Although several approaches have recently been proposed to measure the similarity between business processes, neither the definitions of the similarity notion between processes nor the measure methods have gained wide recognition. In this paper, we define the similarity and the distance based on firing sequences in the context of workflow nets (WF-nets) as the unified reference concepts. However, to many WF-nets, either the number of full firing sequences or the length of a single firing sequence is infinite. Since transition adjacency relations (TARs) can be seen as the genes of the firing sequences which describe transition orders appearing in all possible firing sequences, we propose a practical similarity definition based on the TAR sets of two processes. It is formally shown that the corresponding distance measure between processes is a metric. An algorithm using model reduction techniques for the efficient computation of the measure is also presented. Experimental results involving comparison of different measures on artificial processes and evaluations on clustering real-life processes validate our approach.

Efficient example-based painting and synthesis of 2D directional texture

We present a new method for converting a photo or image to a synthesized painting following the painting style of an example painting. Treating painting styles of brush strokes as sample textures, we reduce the problem of learning an example painting to a texture synthesis problem. The proposed method uses a hierarchical patch-based approach to the synthesis of directional textures. The key features of our method are: 1) Painting styles are represented as one or more blocks of sample textures selected by the user from the example painting; 2) image segmentation and brush stroke directions defined by the medial axis are used to better represent and communicate shapes and objects present in the synthesized painting; 3) image masks and a hierarchy of texture patches are used to efficiently synthesize high-quality directional textures. The synthesis process is further accelerated through texture direction quantization and the use of Gaussian pyramids. Our method has the following advantages: First, the synthesized stroke textures can follow a direction field determined by the shapes of regions to be painted. Second, the method is very efficient; the generation time of a synthesized painting ranges from a few seconds to about one minute, rather than hours, as required by other existing methods, on a commodity PC. Furthermore, the technique presented here provides a new and efficient solution to the problem of synthesizing a 2D directional texture. We use a number of test examples to demonstrate the efficiency of the proposed method and the high quality of results produced by the method.

Reconstruction of curved solids from engineering drawings

This paper presents a new approach for reconstructing solids with planar, quadric and toroidal surfaces from three-view engineering drawings. By applying geometric theory to 3-D reconstruction, our method is able to remove restrictions placed on the axes of curved surfaces by existing methods. The main feature of our algorithm is that it combines the geometric properties of conics with affine properties to recover a wider range of 3-D edges. First, the algorithm determines the type of each 3-D candidate conic edge based on its projections in three orthographic views, and then generates that candidate edge using the conjugate diameter method. This step produces a wire-frame model that contains all candidate vertices and candidate edges. Next, a maximum turning angle method is developed to find all the candidate faces in the wire-frame model. Finally, a general and efficient searching technique is proposed for finding valid solids from the candidate faces; the technique greatly reduces the searching space and the backtracking incidents. Several examples are given to demonstrate the efficiency and capability of the proposed algorithm.

Direct manipulation of FFD: efficient explicit solutions and decomposible multiple point constraints

The ability to directly manipulate an embedded object in the free-form deformation (FFD) method improves controllability. However, the existing solution to this problem involves a pseudo-inverse matrix that requires complicated calculations. This paper solves the problem using a constrained optimization method. We derive the explicit solutions for deforming an object which is to pass through a given target point. For constraints with multiple target points, the proposed solution also involves simple calculations, only requiring solving a system of linear equations. We show that the direct manipulations exhibit the commutative group property, namely commutative, associative, and invertible properties, which further enhance the controllability of FFD. In addition, we show that multiple point constraints can be decomposed into separate manipulations of single point constraints, thus providing the user the freedom of specifying the constraints in any appropriate order.

Constraint Based Region Matching for Image Retrieval

Objects and their spatial relationships are important features for human visual perception. In most existing content-based image retrieval systems, however, only global features extracted from the whole image are used. While they are easy to implement, they have limited power to model semantic-level objects and spatial relationship. To overcome this difficulty, this paper proposes a constraint-based region matching approach to image retrieval. Unlike existing region-based approaches where either individual regions are used or only first-order constraints are modeled, the proposed approach formulates the problem in a probabilistic framework and simultaneously models both first-order region properties and second-order spatial relationships for all the regions in the image. Specifically, in this paper we present a complete system that includes image segmentation, local feature extraction, first- and second-order constraints, and probabilistic regionweight estimation. Extensive experiments have been carried out on a large heterogeneous image collection with 17,000 images. The proposed approach achieves significantly better performance than the state-of-the-art approaches.