• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.6
• /
• pp.2806-2825
• /
• 2018

Unsupervised learning has shown good performance on image, video and audio classification tasks, and much progress has been made so far. It studies how systems can learn to represent particular input patterns in a way that reflects the statistical structure of the overall collection of input patterns. Many promising deep learning systems are commonly trained by the greedy layerwise unsupervised learning manner. The performance of these deep learning architectures benefits from the unsupervised learning ability to disentangling the abstractions and picking out the useful features. However, the existing unsupervised learning algorithms are often difficult to train partly because of the requirement of extensive hyperparameters. The tuning of these hyperparameters is a laborious task that requires expert knowledge, rules of thumb or extensive search. In this paper, we propose a simple and effective unsupervised feature learning algorithm for image classification, which exploits an explicit optimizing way for population and lifetime sparsity. Firstly, a sparse target matrix is built by the competitive rules. Then, the sparse features are optimized by means of minimizing the Euclidean norm ($L_2$) error between the sparse target and the competitive layer outputs. Finally, a classifier is trained using the obtained sparse features. Experimental results show that the proposed method achieves good performance for image classification, and provides discriminative features that generalize well.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.11
• /
• pp.5624-5638
• /
• 2016

A novel method based on Surf points is proposed to detect and lock-track single ground target in aerial videos. Videos captured by moving cameras contain complex motions, which bring difficulty in moving object detection. Our approach contains three parts: moving target template detection, search area estimation and target tracking. Global motion estimation and compensation are first made by grids-sampling Surf points selecting and matching. And then, the single ground target is detected by joint spatial-temporal information processing. The temporal process is made by calculating difference between compensated reference and current image and the spatial process is implementing morphological operations and adaptive binarization. The second part improves KALMAN filter with surf points scale information to predict target position and search area adaptively. Lastly, the local Surf points of target template are matched in this search region to realize target tracking. The long-term tracking is updated following target scaling, occlusion and large deformation. Experimental results show that the algorithm can correctly detect small moving target in dynamic scenes with complex motions. It is robust to vehicle dithering and target scale changing, rotation, especially partial occlusion or temporal complete occlusion. Comparing with traditional algorithms, our method enables real time operation, processing $520{\times}390$ frames at around 15fps.

• International Journal of Contents
• /
• v.11 no.3
• /
• pp.47-53
• /
• 2015

Visual object tracking is a fundamental problem in the field of computer vision, as it needs a proper model to account for drastic appearance changes that are caused by shape, textural, and illumination variations. In this paper, we propose a feature-based visual-object-tracking method with a sparse representation. Generally, most appearance-based models use the gray-scale pixel values of the input image, but this might be insufficient for a description of the target object under a variety of conditions. To obtain the proper information regarding the target object, the following combination of features has been exploited as a corresponding representation: First, the features of the target templates are extracted by using the HOG (histogram of gradient) and LBPs (local binary patterns); secondly, a feature-based sparsity is attained by solving the minimization problems, whereby the target object is represented by the selection of the minimum reconstruction error. The strengths of both features are exploited to enhance the overall performance of the tracker; furthermore, the proposed method is integrated with the particle-filter framework and achieves a promising result in terms of challenging tracking videos.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.5
• /
• pp.2590-2606
• /
• 2017

Generally, research of classical image classification algorithms assume that training data and testing data are derived from the same domain with the same distribution. Unfortunately, in practical applications, this assumption is rarely met. Aiming at the problem, a domain adaption image classification approach based on multi-sparse representation is proposed in this paper. The existences of intermediate domains are hypothesized between the source and target domains. And each intermediate subspace is modeled through online dictionary learning with target data updating. On the one hand, the reconstruction error of the target data is guaranteed, on the other, the transition from the source domain to the target domain is as smooth as possible. An augmented feature representation produced by invariant sparse codes across the source, intermediate and target domain dictionaries is employed for across domain recognition. Experimental results verify the effectiveness of the proposed algorithm.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.12
• /
• pp.3976-3990
• /
• 2022

Currently, cross-attention is widely used in Siamese trackers to replace traditional correlation operations for feature fusion between template and search region. The former can establish a similar relationship between the target and the search region better than the latter for robust visual object tracking. But existing trackers using cross-attention only focus on rich semantic information of high-level features, while ignoring the appearance information contained in low-level features, which makes trackers vulnerable to interference from similar objects. In this paper, we propose a Multi-level Cross-attention Siamese network(MCSiam) to aggregate the semantic information and appearance information at the same time. Specifically, a multi-level cross-attention module is designed to fuse the multi-layer features extracted from the backbone, which integrate different levels of the template and search region features, so that the rich appearance information and semantic information can be used to carry out the tracking task simultaneously. In addition, before cross-attention, a target-aware module is introduced to enhance the target feature and alleviate interference, which makes the multi-level cross-attention module more efficient to fuse the information of the target and the search region. We test the MCSiam on four tracking benchmarks and the result show that the proposed tracker achieves comparable performance to the state-of-the-art trackers.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.9
• /
• pp.2458-2482
• /
• 2023

Medical image segmentation techniques based on convolution neural networks indulge in feature extraction triggering redundancy of parameters and unsatisfactory target localization, which outcomes in less accurate segmentation results to assist doctors in diagnosis. In this paper, we propose a multi-level semantic-rich encoding-decoding network, which consists of a Pooling-Conv-Former (PCFormer) module and a Cbam-Dilated-Transformer (CDT) module. In the PCFormer module, it is used to tackle the issue of parameter explosion in the conservative transformer and to compensate for the feature loss in the down-sampling process. In the CDT module, the Cbam attention module is adopted to highlight the feature regions by blending the intersection of attention mechanisms implicitly, and the Dilated convolution-Concat (DCC) module is designed as a parallel concatenation of multiple atrous convolution blocks to display the expanded perceptual field explicitly. In addition, MultiHead Attention-DwConv-Transformer (MDTransformer) module is utilized to evidently distinguish the target region from the background region. Extensive experiments on medical image segmentation from Glas, SIIM-ACR, ISIC and LGG demonstrated that our proposed network outperforms existing advanced methods in terms of both objective evaluation and subjective visual performance.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.35C no.2
• /
• pp.71-79
• /
• 1998

In this paper, we propose a new mART(modified ART) neural network by combining the winner neuron definition method of SOM(self-organizing map) and the real-time adaptive clustering function of ART(adaptive resonance theory) and construct it in a modular structure, for the purpose of organizing the feature maps of three dimensional targets. Being constructed in a modular structure, the proposed modular mART can effectively prevent the clusters from representing multiple classes and can be trained to organze two dimensional distortion invariant feature maps so as to recognize targets with three dimensional distortion. We also present the recognition result and self-organization perfdormance of the proposed modular mART neural network after carried out some experiments with 14 tank and fighter target models.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.40 no.3
• /
• pp.203-216
• /
• 2003

In this paper a new feature extraction and target classification method is proposed for the recognition part of FLIR(Forwar Looking Infrared)-image-based ATR system. Proposed feature extraction method is "cluster(=set of classes)-based"version of previous fisherfaces method that is known by its robustness to illumination changes in face recognition. Expecially introduced class clustering and cluster-based projection method maximizes the performance of fisherfaces method. Proposed target image classification method is based on the mixture of experts model which consists of RBF-type experts and MLP-type gating networks. Mixture of experts model is well-suited with ATR system because it should recognizee various targets in complexed feature space by variously mixed conditions. In proposed classification method, one expert takes charge of one cluster and the separated structure with experts reduces the complexity of feature space and achieves more accurate local discrimination between classes. Proposed feature extraction and classification method showed distinguished performances in recognition test with customized. FLIR-vehicle-image database. Expecially robustness to pixelwise sensor noise and un-wanted intensity variations was verified by simulation.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.22 no.2
• /
• pp.207-214
• /
• 2019

This paper proposes a new feature extraction method for automatically estimating the number of target and detecting the chaff using high range resolution profile(HRRP). Feature of one-dimensional range profile is expected to be limited or missing due to lack of information according to the time. The proposed method considers the dynamic movements of targets depending on the radial velocity. The observed HRRP sequence is used to construct a time-range distribution matrix, then assuming diverse radial velocities reflect the number of target and seduction chaff launch, the proposed method utilizes the characteristic of the gradient distribution on the time-range distribution matrix image, which is validated by electromagnetic computation data and dynamic simulation.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.05a
• /
• pp.455-457
• /
• 2020

Recent studies demonstrate that deep learning model is easily biased by trained with unbalanced datasets. For example, the deep network can be trained to make a prediction by background feature instead the real target's feature. For those problem, a measurement called leakage was introduced to digitize this tendency. In this paper, we propose augmentation strategy which are used generally in computer vision problem to remedy this bias problem and we showed a simple augmentation methods have a effect to this task with experiments.