Dong-Seok Jeong

Object tracking method using back-projection of multiple color histogram models

Automated object tracking system is needed for unmanned observing and proper recording of important places. In this paper, we propose an object tracking method which uses back-projection of color histogram with multiple models. We can make some representative models of an object from its color histogram distribution. 3D Labeling is introduced to eliminate unsuitable histogram blobs and color histogram models are composed from survived blobs. The position of interested object could be estimated with the back-projection image of each model. The proposed method can reduce the miss tracking even if object enters similar colored region in complex video scenes.

Unified approach for early-phase image understanding using a general decision criterion

Two types of approaches for computer vision are combined to model images or portions thereof, parametrically. These approaches, namely those based on polynomial models and those based on random-field models, are combined based on a general decision criterion for dealing with a variety of modeling strategies. Selection among alternative model structures is in accordance with the tradeoff between sample size and model complexity. Experiments with synthesized images and natural images such as Brodatz textures illustrate some identification and segmentation uses of this unified approach. The implemented segmentation algorithm achieves early-phase region extraction without relying on any contextual or high-level assumptions. A natural result of this is a list of regions, suitable as input for higher-level stages of image understanding in addition to a pixel-labeled image. >

Very fast concentric circle partition-based replica detection method

Image replica detection becomes very active research field recently as the electronic device such as the digital camera which generates digital images spreads out rapidly. As huge amount of digital images leads to severe problems like copyright protection, the necessity of replica detection system gets more and more attention. In this paper, we propose a new fast image replica detector based on concentric circle partition method. The proposed algorithm partitions image into concentric circle with fixed angle from image center position outwards. From these partitioned regions, total of four features are extracted. They are average intensity distribution and its difference, symmetrical difference distribution and circular difference distribution in bitstring type. To evaluate the performance of the proposed method, pairwise independence test and accuracy test are applied. We compare the duplicate detection performance of the proposed algorithm with that of the MPEG-7 visual descriptors. From experimental results, we can tell that the proposed method shows very high matching speed and high accuracy on the detection of replicas which go through many modification from the original. Because we use the hash code as the image signature, the matching process needs very short computation time. And the proposed method shows 97.6% accuracy on average under 1 part per million false positive rate.

Fast panoramic image generation method using morphological corner detection

We present a method of building a panoramic image from adjacent images. The panoramic image is constructed using several images taken by adjacent cameras or parsed from a video, and used for photogrammetry or many computer graphic applications. The perspective transformation, which is estimated from the appropriate corresponding pairs of images, can be used to construct the panoramic image without unwarranted distortion. We used the comer points for the corresponding features, and morphological structures were utilized for fast and robust comer detection. We used the criterion of the comer strength, which guarantees the robust detection of the comer in most situations. For the transformation, 8 parameters were estimated from perspective equations, and bilinear color blending was used to construct a seamless panoramic video. The experiments showed that the proposed method yields fast results with good quality under various conditions.

Object classification using CNN for video traffic detection system

Recently, a lot of research on the use of big data is made, and this paper was aimed to perform classification experiments using CNN for the detected object collected from traffic detectors. In addition the experimental results were compared with the HOG descriptor that is commonly used in existing pedestrian and object classification and wavelet, texture and descriptor that are used in the road surface condition classification. According to the results after applied to the collected RVFTe-10 data, the performances of HOG SVM and CNN were excellent by showing 99.9% and 99.5% respectively.

Enhancement of speech/music decision employing GMM for SMV codec

In this paper, we propose a novel approach to improve the performance of speech/music classification for the selectable mode vocoder (SMV) of 3GPP2 using the Gaussian mixture model (GMM) with a minimum classification error (MCE) method. Also, to enhance We first present an effective analysis of the features and the classification method adopted in the conventional SMV. And then feature vectors which are applied to the GMM are selected from relevant parameters of the SMV for the efficient speech/music classification. The performance of the proposed algorithm is evaluated under various conditions and yields better results compared with the conventional scheme of the SMV.

Semi-fragile watermarking algorithm for detection and localization of temper using hybrid watermarking method in MPEG-2 video

In this paper, a novel semi-fragile watermarking adapted to MPEG-2 video is proposed. It is achieved by using the hybrid watermarking method which is implemented by the combination of the robust watermark applied in DCT and spatial domain and the fragile watermark applied in motion vector. The proposed method can not only detect attacks but also localize them. Besides, it can distinguish malicious attacks such as frame dropping or swapping from non-malicious tempers like re-compression. This method satisfies invisibility and does not need the original information for the watermark detection.

Rough edge detection of low contrast images using consequential local variance maxima

This paper suggests a new edge detection method using local variance maxima which can be applied for finding an auto license plate (ALP) with over 98% success rate in a practical situation. We determine whether a certain point belongs to an edge or not by its length of consequential local variance maxima instead of its magnitude. This is a distinguished difference from the other edge detection methods. The proposed method works effectively even for images of low contrast. This method finds edges by a single pass, with line by line scanning that means no preprocessing and small time consumption.

A study on the postprocessing of functional MRI using overcomplete wavelet transform

Presents a new algorithm for detecting activity-dependent signal changes in Functional Magnetic Resonance Imaging (FMRI) image sequences. In this paper, 3-D Multi-Resolution Analysis (MRA) using Overcomplete Wavelet Transform (OWT) is proposed. It is like temporal wavelet thresholding in 1-D MRA, spatial wavelet domain filtering in 2-D MRA and Region Of Interest (ROI) analysis. Reconstructed images by the inverse wavelet transform are tested with Statistical Parametric Mapping (SPM). The results of extensive tests have shown that the proposed algorithm is reliable in both high SNR and low SNR image sequences.

Bag-of-features signature using invariant region descriptor for object retrieval

In recent years, a content-based method such as ‘bag-of-features’ (BoF) is coming to the fore as an object recognition and classification technique. This paper proposes a BoF signature using invariant region descriptor for object retrieval. The region descriptors are extracted from dense sampled regions in the training images. These descriptors are quantized by hierarchical k-means clustering in a vocabulary tree of visual words. Each image is represented by occurrence of visual words, and then we use linear combination distance measure in the matching. In the experiments, we use object images that are taken in different condition to evaluate the retrieval performance. The test results show that the proposed method outperforms the BoF method using SURF descriptor. The proposed method searches 2.9 correct images among 3 on average up to the top 3% rank in database. Therefore, the proposed method is considered as an effective technique in terms of retrieval accuracy.