Tianrui Li

A Learning-Based Framework for Image Segmentation Evaluation

Image segmentation is a fundamental task in automatic image analysis. However, there is still no generally accepted effectiveness measure which is suitable for evaluating the segmentation quality in every application. In this paper, we propose an evaluation framework which benefits from multiple stand-alone measures. To this end, different segmentation evaluation measures are chosen to evaluate segmentation separately, and the results are effectively combined using machine learning methods. We train and implement this framework in the segmentation dataset which contains images of different contents with segmentation ground truth produced by human. In addition, we provide human evaluation of image segmentation pairs to benchmark the evaluation results of the measures. Experimental results show a better performance than the stand-alone methods.

Incremental Updating Rough Approximations in Interval-valued Information Systems

Interval-valued Information System (IvIS) is a generalized model of single-valued information system, in which the attribute values of objects are all interval values instead of single values. The attribute set in IvIS is not static but rather dynamically changing over time with the collection of new information. The rough approximations may evolve accordingly, which should be updated continuously for data analysis based on rough set theory. In this paper, on the basis of the similarity-based rough set theory in IvIS, we first analyze the relationships between the original approximation sets and the updated ones. And then we propose the incremental methods for updating rough approximations when adding and removing attributes, respectively. Finally, a comparative example is used to validate the effectiveness of the proposed incremental methods.

A Learning-Based Framework for Supervised and Unsupervised Image Segmentation Evaluation

Image segmentation is a fundamental task in automatic image analysis. However, there is still no generally accepted effectiveness measure which is suitable for evaluating the segmentation quality in every application. In this paper, we propose an evaluation framework which benefits from multiple stand-alone measures. To this end, different segmentation evaluation measures are chosen to evaluate segmentation separately, and the results are effectively combined using machine learning methods. We train and implement this framework in our brand-new segmentation dataset which contains images of different contents with segmentation ground truth and Weizmann segmentation database (WSD). In addition, we provide human evaluation of image segmentation pairs to benchmark the evaluation results of the measures. Experimental results show a better performance than the stand-alone methods.

Approaches for Updating Approximations in Set-Valued Information Systems While Objects and Attributes Vary with Time

Rough set theory is an important tool for knowledge discovery. The lower and upper approximations are basic operators in rough set theory. Certain and uncertain if-then rules can be unrevealed from different regions partitioned by approximations. In real-life applications, data in the information system are changing frequently, for example, objects, attributes, and attributes’ values in the information system may vary with time. Therefore, approximations may change over time. Updating approximations efficiently is crucial to the knowledge discovery. The set-valued information system is a general model of the information system. In this chapter, we focus on studying principles for incrementally updating approximations in a set-valued information system while attributes and objects are added. Then, methods for updating approximations of a concept in a set-valued information system is given while attributes and objects change simultaneously. Finally, an extensive experimental evaluation verifies the effectiveness of the proposed method.

A Multi-granulation Model under Dominance-Based Rough Set Approach

Multi-granulation rough set is a generalization of Rough Set Theory (RST) in order to adapt to cases that there are multiple relations in the universe. To allow Dominance-based Rough Set Approach (DRSA) being applied in multiple relations cases, we propose a multi-granulation model based on DRSA. A numerical example is employed to validate the rationality and feasibility of our model.

An Incremental Approach for Updating Approximations Based on Set-Valued Ordered Information Systems

Incremental learning is an efficient technique for knowledge discovery in a dynamic database. Rough set theory is an important mathematical tool for data mining and knowledge discovery in information systems. The lower and upper approximations in the rough set theory may change while data in the information system evolves with time. In this paper, we focus on the incremental updating principle for computing approximations in set-valued ordered information systems. The approaches for updating approximations are proposed when the object set varies over time.

Online Evaluation System of Image Segmentation

Over the past few decades, many applications have been developed for quantify the performance of image segmentation algorithms. However, applying standard interactive segmentation software for large image datasets would be an extremely laborious and a time-consuming task. In this chapter, we present an online evaluation framework for analysis and visualization features via Internet communication that serves as a remote system. One of the features of the online evaluation system is the combination of MATLAB and Java to provide benefit platform that create the application with advanced mathematical computation as well as a well-designed graphical interface. This framework provides a web-based tool for validating the efficiency of segmentation algorithms. Experimental results show the possible ways in order to transfer MATLAB algorithm into a MATLAB license-free that make algorithm developed within MATLAB run all cross the platform and available on internet. In addition, the implementation methodology reported can be reused for other similar software engineering tasks.

A Novel Graph Cut Algorithm for Weak Boundary Object Segmentation

Image segmentation plays an important role in high-level visual recognition tasks. In recent years, the combinatorial graph cut algorithm has been successfully applied to image segmentation because it offers numerically robust global minimum. For low-level image segmentation, intensity is a widely used regional cue. However, when comes to weak boundary, it is often not enough to discriminate the object of interest. In this paper, we extend the standard graph cut algorithm by taking into account the gradient direction of neighboring pixels as an additional cue. A new energy function is proposed to fuse the intensity and gradient cues. Experimental results show that our method is more robust and helpful to detect the low-contrast boundaries.

An Incremental Approach for Updating Approximations of Rough Fuzzy Sets under the Variation of the Object Set

The lower and upper approximations are basic concepts in rough set theory, and the approximations will change dynamically over time. Incremental methods for updating approximations in rough set theory and its extension has been received much attention recently. This paper presents an approach for incrementally updating approximations of fuzzy rough sets in dynamic fuzzy decision systems when a single object immigrating and emigrating. Examples are employed to illustrate the proposed approach.