• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.6
• /
• pp.102-111
• /
• 2009

In this paper, we propose a de-interlacing algorithm that combines a motion compensation (MC) method and the vertical-temporal filter with motion compensation (MC V-T filter) according to motion compensation reliability. The MC method represent one of the best ways of improving the resolution of de-interlaced frames, but it may introduce motion compensation artifacts in regions with incorrect motion information. In these regions, the MC V-T filter that is very robust to motion vector errors can be used to correct motion compensation artifacts. The combination between two methods is controlled by the motion compensation reliability that is measured by analyzing the estimated motion vectors and the results of MC. The motion compensation reliability contains information about motion compensation artifacts of MC results and determines the combination weight according to this information. Therefore, the combination rule of the proposed method is more accurate than those of the conventional methods and it enables the proposed method to provide high quality video sequences without producing any visible artifacts. Experimental results with various test sequences show that the proposed algorithm outperforms conventional algorithms in terms of both visual and numerical criteria.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.2
• /
• pp.121-128
• /
• 2003

Acoustic Echo Cancellation(m) is very active research topic having many applications like teleconference and hands-free communication and it employs Double Talk Detector(DTD) to indicate whether the near-end speaker is active or not. However. the DTD is very sensitive to the variation of acoustical environment and it sometimes provides wrong information about the near-end speaker. In this paper, we are focusing on the development of robust DTD algorithm which is a basic building block for reliable AEC system. The proposed AEC system consists of delayless subband AEC and narrow-band DTD. Delayless subband AEC has proven to have excellent performance of echo cancellation with a low complexity and high convergence speed. In addition, it solves the signal delay problem in the existing subband AEC. On the other hand, the proposed narrowband DTD is operating on low frequency subband. It can take most advantages from the narrow subband such as a low computational complexity due to the down-sampling and the reliable DTD decision making procedure because of the low-frequency nature of the subband signal. From the simulation results of the proposed narrowband DTD and wideband DTD, we confirm that the proposed DTD outperforms the wideband DTD in a sense of removing possible false decision making about the near-end speaker activity.

• The KIPS Transactions:PartB
• /
• v.9B no.2
• /
• pp.245-254
• /
• 2002

IC package marking inspection system using machine vision locates and identifies the target elements from input image, and decides the quality of marking by comparing the extracted target elements with the standard patterns. This paper proposes an adaptive multi-level thresholding (AMLT) method which is suitable for a series of operations such as locating the target IC package, extracting the characters, and detecting the Pinl dimple. It also proposes a dynamic matching unit selection (DMUS) method which is robust to noises as well as effective to catch out the local marking errors. The main idea of the AMLT method is to restrict the inputs of Otsu's thresholding algorithm within a specified area and a partial range of gray values. Doing so, it can adapt to the specific domain. The DMUS method dynamically selects the matching unit according to the result of character extraction and layout analysis. Therefore, in spite of the various erroneous situation occurred in the process of character extraction and layout analysis, it can select minimal matching unit in any environment. In an experiment with 280 IC package images of eight types, the correct extracting rate of IC package and Pinl dimple was 100% and the correct decision rate of marking quality was 98.8%. This result shows that the proposed methods are effective to IC package marking inspection.

• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.