Seung Wan Han

A method for counting moving and stationary people by interest point classification

While the methods for people counting based on moving interest points have been shown good performance, the problem related to count static or temporarily stopped people remains particularly challenging. This paper presents a novel method for people counting which is considering moving and stationary people. The proposed method first separates the moving and static points by the motion information. Then, the temporarily static points of stationary people classified by analyzing the texture information of the points between the current and the eigenbackground image. Finally, we estimate the number of people using the moving and the temporarily static points. The experimental results show that the proposed method can identify the static points related to people correctly. Additionally, the results confirm that the proposed method reduces the estimation error in the sequences containing the many temporarily static people.

An Objectionable Image Detection System Based on Region of Interest

In this paper, we propose the objectionable image detection system based on ROI. The proposed system excels in that ROI detection method is specialized in objectionable image detection. In addition, we propose the novel feature consisting of weighted SCD based on ROI and skin color structure descriptor for classifying objectionable image. Using the ROI detection method, we can reduce the noisy information in image and extract more accurate features for classifying objectionable image. We can also increase the performance of classification system using the skin color information which is the most representative characteristic of objectionable image. An objectionable image is classified as follows. First, the proposed system extracts the ROI from an input image, and then generates the feature using the color information of ROI image and the skin color structure information. The generated feature is used as an input to the SVMs which determines whether an image is objectionable or not. Experimental results show that the proposed ROI detection method can detect ROI exactly. Additionally, the proposed system can achieve the better performance than existing system.

New binary descriptors based on BRISK sampling pattern for image retrieval

Over the last decade, feature point descriptors such as SIFT have become indispensable tools in the computer vision community. But, the descriptors high computational overhead becomes a significant concern when it has to be on a device with limited computational and storage resources. In order to make descriptors faster to compute and more compact, several binary descriptors such as ORB and BRISK have been proposed. These binary descriptors are not successful in image retrieval, so we propose new binary descriptors to increase the accuracy while maintaining computational efficiency.

Skin information combined saliency detection for classifying objectionable images

Although visual attention models have shown good performance in general environments, applying it for filtering objectionable image needs consideration for skin information. Therefore, in this paper, we proposed the saliency detection method specialized in adult image classification using the skin color and the contrast information. Skin information combined saliency map is generated by merging the saliency map of visual attention model and the skin probability map. In order to prove the effectiveness of the proposed method, we consider the use of a saliency map in objectionable image classification.

A Hierarchical Text Rating System for Objectionable Documents

In this paper, we classified the objectionable texts into four rates according to their harm- fulness and proposed the hierarchical text rating system for objectionable documents. Since the documents in the same category have similarities in used words, expressions and structure of the document, the text rating system, which uses a single classification model, has low accuracy. To solve this problem, we separate objectionable documents into several subsets by using their properties, and then classify the subsets hierarchically. The proposed system consists of three layers. In each layer, we select features using the chi-square statistics, and then the weight of the features, which is calculated by using the TF-IDF weighting scheme, is used as an input of the non-linear SVM classifier. By means of a hierarchical scheme using the different features and the different number of features in each layer, we can characterize the objectionability of documents more effectively and expect to improve the performance of the rating system. We compared the performance of the proposed system and performance of several text rating systems and experimental results show that the proposed system can archive an excellent classification performance.

Objectionable video classification exploiting cascade classifier

For the image domain, much research has been done to protect children from the harmful exposure to the objectionable images. Although image classification techniques require low computation time, it requires longer computation time to classify the videos. To classify the objectionable videos, it needs to process the many representative key-frames. To reduce the processing time of video classification, this paper proposes an objectionable video classification method exploiting cascade classifier. The experimental results show that the proposed method reduces the processing time without the degradation of classification performance.

Evaluating weighting schemes for adult image detection using bag of visual words

As a growing number of minors are exposed to pornographic multimedia, the need to prevent inappropriate access to this has increased. Adult image detection i s a n important technology for this purpose, and several algorithms have been proposed. However, the simple methods such as analyzing skin regions or global histogram suffer from low classification accuracy. Recent advances in adult image detection used bag of visual words approach and showed promising result. In this paper, we experiment with several weighting schemes for visual words histogram construction and show that classification accuracy can be increased.

Secure rejoining scheme for dynamic sensor networks

The establishment of cryptography keys is one of the challenging problems in the sensor networks. Key Infection [1] is a promising model to solve this problem on the commodity sensor networks without complex mechanism. This model, however, does not consider the mobility of sensors, so if sensor nodes move out of initial communication range, then they cannot rejoin the network. So, key infection model has been limited to the static sensor network. To be applied on the dynamic sensor network, therefore, key infection model has to be extended to handle nodes rejoining. In this paper, we propose secure rejoining scheme for dynamic sensor networks and verify the proposed scheme formally. Our scheme is secure, since it uses old pair-wise key information to verify sensor nodes rejoining. Furthermore, our scheme does not require additional verification information and maintains the reasonable number of links.

Classifying objectionable images based on MPEG-7 visual descriptors and saliency detection

Much research has been done to block objectionable images by analyzing the visual contents. The region-of-interest based approach are more accurate in describing the image content than the approach using the whole image in the field of content-based image retrieval. Therefore, this paper proposes a method for classifying objectionable images based on global features and salient region. By using the features concatenated the visual descriptors extracted from whole image and salient region, the proposed method can accurately describe the characteristic of objectionable image. The experimental results show that the proposed method increases the performance of objectionable image classification.

Fast image retrieval with grid-based keypoint detector and binary descriptor.

As an alternative to vector-based descriptors, such as SIFT and SURF, more computationally efficient binary descriptors, such as BRISK and ORB, have recently been proposed. These binary descriptors are usually used in combination with a novel scale-space FAST-based detector to be suitable for real-time applications, but it consumes more time than creating binary descriptors. Therefore, if accuracy can be kept similar, keypoint sampling by a grid is better than FAST-based detector because it consumes almost no time. In this paper, grid-based sampling and BRISK keypoint detector are tested for image retrieval. Experimental results demonstrate that grid-based sampling out performs keypoint detector in terms of accuracy and processing speed.