• Journal of the Korea Academia-Industrial cooperation Society
• /
• 제8권6호
• /
• pp.1524-1529
• /
• 2007

In this paper, automatic shot panorama construction method is presented. For construction of panorama image, conventional panoramic techniques manually took two panorama members, but the proposed method automatically takes panorama members according to moving camera and constructs panorama image. The panorama members are automatically selected and taken by tracking region over image stream form camera. Matching region for panorama including the tracking region in the members is selected and applied by invariant feature panoramic method. Our method can automatically shot panorama members and has merit of high processing speed. In the experiments, it was shown that the algorithm required about 0.89 second in processing time, about two times shorter than existing invariant feature based one(6), for color images of $320{\times}240$ size.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• 제48권2호
• /
• pp.66-77
• /
• 2011

We propose a framework for grouping photographic advertising images that employs a hierarchical indexing scheme based on efficient feature combinations. The study provides one specific application of effective tools for monitoring photographic advertising information through online and offline channels. Specifically, it develops a preprocessor for advertising image information tracking. We consider both global features that contain general information on the overall image and local features that are based on local image characteristics. The developed local features are invariant under image rotation and scale, the addition of noise, and change in illumination. Thus, they successfully achieve reliable matching between different views of a scene across affine transformations and exhibit high accuracy in the search for matched pairs of identical images. The method works with global features in advance to organize coarse clusters that consist of several image groups among the image data and then executes fine matching with local features within each cluster to construct elaborate clusters that are separated by identical image groups. In order to decrease the computational time, we apply a conventional clustering method to group images together that are similar in their global characteristics in order to overcome the drawback of excessive time for fine matching time by using local features between identical images.

• Journal of Platform Technology
• /
• 제6권4호
• /
• pp.26-33
• /
• 2018

In this paper, we propose an image retrieval algorithm for real-time processing and design it as hardware. The proposed method is based on the classification of BoWs(Bag of Words) algorithm and proposes an image search algorithm using bit stream. K-fold cross validation is used for the verification of the algorithm. Data is classified into seven classes, each class has seven images and a total of 49 images are tested. The test has two kinds of accuracy measurement and speed measurement. The accuracy of the image classification was 86.2% for the BoWs algorithm and 83.7% the proposed hardware-accelerated software implementation algorithm, and the BoWs algorithm was 2.5% higher. The image retrieval processing speed of BoWs is 7.89s and our algorithm is 1.55s. Our algorithm is 5.09 times faster than BoWs algorithm. The algorithm is largely divided into software and hardware parts. In the software structure, C-language is used. The Scale Invariant Feature Transform algorithm is used to extract feature points that are invariant to size and rotation from the image. Bit streams are generated from the extracted feature point. In the hardware architecture, the proposed image retrieval algorithm is written in Verilog HDL and designed and verified by FPGA and Design Compiler. The generated bit streams are stored, the clustering step is performed, and a searcher image databases or an input image databases are generated and matched. Using the proposed algorithm, we can improve convenience and satisfaction of the user in terms of speed if we search using database matching method which represents each object.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• 제30B권9호
• /
• pp.64-75
• /
• 1993

This paper proposes a recovering technique of the elevation map by stereo modeling of the aerial image sequence which is transformed based on the aircraft situation. The area-based stereo matching method is simulated and the various parameters are experimentally chosen. In a depth extraction step, the depth is determined by solving the vector equation. The equation is suitable for stereo modeling of aerial images which do not satisfy the epipolar constraint. Also, the performance of the conventional feature-based matching scheme is compared. Finally, techniques analyzing the accuracy of the recovered elevation map (REM) are described. The analysis includes the error estimation for both height and contour lines, where the accuracy is based on the measurements of deviations from the estimates obtained manually. The experimental results show the efficiency of the proposed technique.

• ETRI Journal
• /
• 제32권6호
• /
• pp.871-880
• /
• 2010

Many applications dealing with image management need a technique for removing duplicate images or for grouping related (near-duplicate) images in a database. This paper proposes a concentric circle-based image signature which makes it possible to detect near-duplicates rapidly and accurately. An image is partitioned by radius and angle levels from the center of the image. Feature values are calculated using the average or variation between the partitioned sub-regions. The feature values distributed in sequence are formed into an image signature by hash generation. The hashing facilitates storage space reduction and fast matching. The performance was evaluated through discriminability and robustness tests. Using these tests, the particularity among the different images and the invariability among the modified images are verified, respectively. In addition, we also measured the discriminability and robustness by the distribution analysis of the hashed bits. The proposed method is robust to various modifications, as shown by its average detection rate of 98.99%. The experimental results showed that the proposed method is suitable for near-duplicate detection in large databases.

• Journal of the Semiconductor & Display Technology
• /
• 제12권1호
• /
• pp.1-7
• /
• 2013

Grey-level statistical models have been widely used in many applications for object location and identification. However, conventional models yield some problems in model refinement when training images are not properly aligned, and have difficulties for real-time recognition of arbitrarily rotated models. This paper presents improved grey-level statistical models that align training images using image or feature matching to overcome problems in model refinement of conventional models, and that enable real-time recognition of arbitrarily rotated objects using efficient hierarchical search methods. Edges or features extracted from a mean training image are used for accurate alignment of models in the search image. On the aligned position and orientation, fitness measure based on grey-level statistical models is computed for object recognition. It is demonstrated in various experiments in PCB inspection that proposed methods are superior to conventional methods in recognition accuracy and speed.

• Journal of Industrial Technology
• /
• 제28권B호
• /
• pp.143-148
• /
• 2008

A real-time face detection is to find human faces robustly under the cluttered background free from the effect of occlusion by other objects or various lightening conditions. We propose a face detection system for real-time applications using wavelet decomposition method based on Gabor features. Firstly, skin candidate regions are extracted from the given image by skin color filtering and projection method. Then Gabor-feature based template matching is performed to choose face cadidate from the skin candidate regions. The chosen face candidate region is transformed into 2-level wavelet decomposition images, from which feature vectors are extracted for classification. Based on the extracted feature vectors, the face candidate region is finally classified into either face or nonface class by the Levenberg-Marguardt back-propagation neural network.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제18권2호
• /
• pp.311-326
• /
• 2024

The rapid development of neural network technology promotes the neural network model driven by big data to overcome the texture effect of complex objects. Due to the limitations in complex scenes, it is necessary to establish custom template matching and apply it to the research of many fields of computational vision technology. The dependence on high-quality small label sample database data is not very strong, and the machine learning system of deep feature connection to complete the task of texture effect inference and speculation is relatively poor. The style transfer algorithm based on neural network collects and preserves the data of patterns, extracts and modernizes their features. Through the algorithm model, it is easier to present the texture color of patterns and display them digitally. In this paper, according to the texture effect reasoning of custom template matching, the 3D visualization of the target is transformed into a 3D model. The high similarity between the scene to be inferred and the user-defined template is calculated by the user-defined template of the multi-dimensional external feature label. The convolutional neural network is adopted to optimize the external area of the object to improve the sampling quality and computational performance of the sample pyramid structure. The results indicate that the proposed algorithm can accurately capture the significant target, achieve more ablation noise, and improve the visualization results. The proposed deep convolutional neural network optimization algorithm has good rapidity, data accuracy and robustness. The proposed algorithm can adapt to the calculation of more task scenes, display the redundant vision-related information of image conversion, enhance the powerful computing power, and further improve the computational efficiency and accuracy of convolutional networks, which has a high research significance for the study of image information conversion.

• Journal of the Korea Society of Computer and Information
• /
• 제28권4호
• /
• pp.31-39
• /
• 2023

In this paper, we propose an Internal/External Knowledge Distillation (IEKD), which utilizes both external correlations between feature maps of heterogeneous models and internal correlations between feature maps of the same model for transferring knowledge from a teacher model to a student model. To achieve this, we transform feature maps into a sequence format and extract new feature maps suitable for knowledge distillation by considering internal and external correlations through a transformer. We can learn both internal and external correlations by distilling the extracted feature maps and improve the accuracy of the student model by utilizing the extracted feature maps with feature matching. To demonstrate the effectiveness of our proposed knowledge distillation method, we achieved 76.23% Top-1 image classification accuracy on the CIFAR-100 dataset with the "ResNet-32×4/VGG-8" teacher and student combination and outperformed the state-of-the-art KD methods.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• 제34S권9호
• /
• pp.109-117
• /
• 1997

A new relaxation algorithm based on distribution of matched errors and possibility is proposed to solve efficiently correspondence problem. This algorithm can be applied to various method, such as BMA, feature-, and region-based matching methods, by modifying its smoothness function. It consists of two stages which are transformation and iteration process. In transformation stage, the errors obtained by any matching algorithm are transformed to possibility values according to these statistical distribution. Each grade of possility is updated by some constraints which are defined as smoothness, uniqueness, and discontinuity factor in iteration stage. The discontinuity factor is used to reserve discontinuity of disparity. In conventional methods, it is difficult to find proper weights and stop condition, because only two factors, smoothness and uniqueness, have been used. However, in the proposed mthod, the more smoothing is not ocurred because of discontinuity factor. And it is efective to the various image, even if the image has a severe noise and repeating patterns. In addition, it is shown that the convergence rate and the quality of output are improved.