• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.5
• /
• pp.1433-1449
• /
• 2023

In recent years, the rapid development of social networks has led to a rapid increase in the amount of information available on the Internet, which contains a large amount of sensitive information related to pornography, politics, and terrorism. In the aspect of sensitive image detection, the existing machine learning algorithms are confronted with problems such as large model size, long training time, and slow detection speed when auditing and supervising. In order to detect sensitive images more accurately and quickly, this paper proposes a multiclassification sensitive image detection method based on lightweight Convolutional Neural Network. On the basis of the EfficientNet model, this method combines the Ghost Module idea of the GhostNet model and adds the SE channel attention mechanism in the Ghost Module for feature extraction training. The experimental results on the sensitive image data set constructed in this paper show that the accuracy of the proposed method in sensitive information detection is 94.46% higher than that of the similar methods. Then, the model is pruned through an ablation experiment, and the activation function is replaced by Hard-Swish, which reduces the parameters of the original model by 54.67%. Under the condition of ensuring accuracy, the detection time of a single image is reduced from 8.88ms to 6.37ms. The results of the experiment demonstrate that the method put forward has successfully enhanced the precision of identifying multi-class sensitive images, significantly decreased the number of parameters in the model, and achieved higher accuracy than comparable algorithms while using a more lightweight model design.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.28 no.3
• /
• pp.271-280
• /
• 2004

Assessment of image registration for Pressure Sensitive Paint (PSP) was performed. A 16 bit camera and LED lamp were used with Uni-FIB paint (ISSI). Because of model displacement and deformation at 'wind-on' condition, a large error of the intensity ratio was induced between 'wind-on' and' wind-off images. To correct the error, many kinds of image registrations were tested. At first, control points were marked on the model surface to find the coefficients of polynomial transform functions between the 'wind-off' 'wind-on' images. The 2nd-order polynomial function was sufficient for representing the model displacement and deformation. An automatic detection scheme was introduced to find the exact coordinates of the control points. The present automatic detection algorithm showed more accurate and user-friendly than the manual detection algorithm. Since the coordinates of transformed pixel were not integer, five interpolation methods were applied to get the exact pixel intensity after transforming the 'wind-on' image. Among these methods, the cubic convolution interpolation scheme gave the best result.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2003.10a
• /
• pp.999-1002
• /
• 2003

In this study, embodied supervisory system that apply motion detection technique to small web camera and detects watch picture. Motion detection technique that use pixel value of car image that use in existing need memory to store background image. Also, there is sensitive shortcoming at increase of execution time by data process of pixel unit and noise. Suggested technique that compare extracting motion information by block unit to do to have complexion that solve this shortcoming and is strong at noise. Because motion information by block compares block characteristic value of image without need frame memory, store characteristic cost by block of image. Also, can get effect that reduce influence about noise and is less sensitive to flicker etc.. of camera more than motion detection that use pixel value in process that find characteristic value by block unit.

• Journal of Information Processing Systems
• /
• v.14 no.6
• /
• pp.1331-1346
• /
• 2018

Aiming at the problem that the gradient-based edge detection operators are sensitive to the noise, causing the pseudo edges, a triqubit-state measurement-based edge detection algorithm is presented in this paper. Combing the image local and global structure information, the triqubit superposition states are used to represent the pixel features, so as to locate the image edge. Our algorithm consists of three steps. Firstly, the improved partial differential method is used to smooth the defect image. Secondly, the triqubit-state is characterized by three elements of the pixel saliency, edge statistical characteristics and gray scale contrast to achieve the defect image from the gray space to the quantum space mapping. Thirdly, the edge image is outputted according to the quantum measurement, local gradient maximization and neighborhood chain code searching. Compared with other methods, the simulation experiments indicate that our algorithm has less pseudo edges and higher edge detection accuracy.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.8 no.1
• /
• pp.74-79
• /
• 2004

In this study, embodied supervisory system that apply motion detection technique to small web camera and detects watch picture. Motion detection technique that use pixel value of car image that use in existing need memory to store background image. Also, there is sensitive shortcoming at increase of execution time by data process of pixel unit and noise. Suggested technique that compare extracting motion information by block unit to do to have complexion that solve this shortcoming and is strong at noise. Because motion information by block compares block characteristic value of image without need frame memory, store characteristic cost by block of image. Also, can get effect that reduce influence about noise and is less sensitive to flicker etc.. of camera more than motion detection that use pixel value in process that find characteristic value by block unit.

• Proceedings of the KIEE Conference
• /
• 1995.11a
• /
• pp.225-227
• /
• 1995

An object detection algorithm using a modified IDM(Image Differential Method) is proposed for detecting an object in a level crossing area. The conventional object detection method using LASER light has the deadzone that it cannot detect small objects, while the object detection method using image data in a level crossing area can detect such small objects. But the image data in a level crossing area can be changeable easily because the data is outdoor and sensitive to such surrounding environments as the change of the sun beam, the shadow of cars, and so on. So we resolve these problems by adding the normalization and the process for shadow of the image data in a level crossing area to the basic IDM(Image Differential Method).

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.4
• /
• pp.294-298
• /
• 2009

In this study, we proposes the method to improve the quality of the image through the edge extraction more delicately. Our method is named ESII(Edge Sensitive Image Interpolation) and doesn't use the fixed parameter of the interpolation kernel. However, it changes the parameter of pixel which is interpolated to the high definition image using the proper information from the surrounding pixels. It reconstructs the image by using the LSE(Least Square Error) and determining the pixel values to make the CME(Camera Modelling Error) minimized. Compared to the conventional methods, suggested method shows the higher quality of subjective and objective image definition and lessons the computational complexity by separating the image into 1-D data.

• Journal of the Semiconductor & Display Technology
• /
• v.22 no.4
• /
• pp.108-112
• /
• 2023

Precision manufacturing technology is rapidly developing due to the extreme miniaturization of semiconductor processes to comply with Moore's Law. Accurate and precise alignment, which is one of the key elements of the semiconductor pre-process and post-process, is very important in the semiconductor process. The center detection of wafer align marks plays a key role in improving yield by reducing defects and research on accurate detection methods for this is necessary. Methods for accurate alignment using traditional image sensors can cause problems due to changes in image brightness and noise. To solve this problem, engineers must go directly into the line and perform maintenance work. This paper emphasizes that the development of AI technology can provide innovative solutions in the semiconductor process as high-resolution image and image processing technology also develops. This study proposes a new wafer center detection method through variable thresholding. And this study introduces a method for detecting the center that is less sensitive to the brightness of LEDs by utilizing a high-performance object detection model such as YOLOv8 without relying on existing algorithms. Through this, we aim to enable precise wafer focus detection using artificial intelligence.

• Journal of Sensor Science and Technology
• /
• v.31 no.2
• /
• pp.131-137
• /
• 2022

Crimes related to illegal cameras are steadily increasing and causing social problems. Owing to the development of camera technology, the miniaturization and high performance of illegal cameras have caused anxiety among many people. This study is for detecting hidden cameras effectively such that they could not be easily detected by human eyes. An image sensor-based module with 940 nm wavelength infrared detection technology was developed, and an image processing algorithm was developed to selectively detect illegal cameras. Based on the Android smartphone environment, image processing technology was applied to an image acquired from an infrared camera, and a detection sensor module that is less sensitive to ambient brightness noise was studied. Experiments and optimization studies were conducted according to the Gaussian blur size, adaptive threshold size, and detection distance. The performance of the infrared image-based illegal camera detection sensor module was excellent. This is expected to contribute to the prevention of crimes related to illegal cameras.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2009.01a
• /
• pp.207-210
• /
• 2009

Edge detection is an important process in low level image processing. But many proposed methods for edge detection are not very robust to the image noise and are not flexible for different images. To solve the both problems, an algorithm is proposed which eliminate the noise by mean shift algorithm in advance, and then adaptively determine the double thresholds based on gradient histogram and minimum interclass variance, With this algorithm, it can fade out almost all the sensitive noise and calculate the both thresholds for different images without necessity to setup any parameter artificially, and choose edge pixels by fuzzy algorithm.