Jo Aug 29, 2023
Nowadays, consumer mobile phones come in with ultra-high resolution cameras, and an incredible number of high-resolution images and videos are created every day. The images and videos have to be downscaled with very large factors to be displayed on general screens.
Deep learning-based downscaling methods show superior performance only for some predetermined integer factors such as 2, 3 and 4. For arbitrary factors, the latest image downscaling algorithms preserve edges and fine details but still suffer from noise amplification. They make undesirable artifacts especially when a downscaling factor is very large.
Kim Su Hyon, a researcher at the Faculty of Information Science and Technology, has proposed an algorithm referred to as NDPID (Noise-free DPID or New DPID) for downscaling ultra-high resolution images to a thumbnail size in real-time without amplifying noise. The proposed algorithm is based on inverse joint bilateral filtering using an area pixel model and moving average.
Unlike the DPID, which employs a rectangular function (box filtering) as the spatial kernel, the NDPID uses two-step 1D APID (Area Pixel model based Image Downscaling) filter. The main reason for employing this 1D spatial kernel is to decompose the proposed downscaling algorithm into two subsequent processes each of which performs capturing pixels’ distinctness for their weights and smoothing of the weights. By these two processes, the algorithm alleviates an isolated noise pixel twice but a thin line (important detail) only once. Consequently, the lines and edges survive while the NDPID alleviates the isolated noise pixel in both horizontal and vertical smoothing processes.
The proposed algorithm is much faster than state-of-the-art downscalers and is free from the restraints of predetermined integer downscaling factors. The experimental results show that the proposed algorithm is about 7.37% faster on average than the DPID, the fastest detail-preserving image downscaler in use. GPU implementation of the algorithm downscales a 2K video to 128-pixel width without temporal artifacts at the speed of 116 frames per second. Moreover, the PSNR and SSIM scores achieved by his method were respectively 35.9% and 16.5% higher on average than the highest values scored by the existing methods when downscaling images contaminated by 5% salt and pepper noise.
If further information is needed, please refer to his paper “A New Rapid and Detail-Preserving Image Downscaling Without Noise Amplification” in “IEEE Access” (SCI).
Jo Aug 28, 2023
Conventional contrast enhancement algorithms such as histogram equalization and adaptive gamma correction overemphasize images without preserving edges, lose details near edges or amplify noise. These are more obvious in case of enhancing dark images with non-uniform illumination, which contain both bright and dark regions. This is mainly because Histogram Equalization and adaptive gamma correction perform a global transformation considering only pixel brightness level.
Kim Thae Song, a section head at the Faculty of Information Science and Technology, has proposed a new algorithm to enhance dark image contrast with non-uniform illumination preserving edges.
His contributions are as follows.
He defined a new edge intensity histogram which represents local brightness variations for each brightness level. With the edge intensity histogram instead of the luminance histogram, adjusting dynamic range adaptively, he obtained a transformation function for efficient contrast enhancement which preserves edges, details and naturalness. To avoid noise amplification, he decomposed an input image into a base and detailed layer and calculated the edge intensity histogram for only the base layer. He enhanced only the base layer using the edge intensity histogram and combined it with the detailed layer which is linearly transformed.
He compared the performance of his method with existing methods such as HE, GMHE, EGEHE, AGCWD, AGCWHD and LTH using a set of dark images taken from the Caltech-256, NCEA, Kodak. To evaluate the proposed method, he used various metrics such as entropy, MSSIM, GMSD, EBCM and AMBE.
The experimental results showed that the proposed method is very effective for enhancing dark images with non-uniform illumination.
If further information is needed, please refer to his paper “An improved contrast enhancement for dark images with non-uniform illumination based on edge preservation” in “Multimedia Systems” (SCI).
Jo Aug 26, 2023
With a flood of display devices with different resolutions, there are increasing demands for efficient image scaling algorithms, which can improve performances of real-time applications such as remote desktop and screen sharing. Image downscaling algorithms can be divided into two classes ― content-adaptive and non-adaptive. Non-adaptive algorithms produce aliasing, blur and halo artifacts though they are very fast. Content-adaptive algorithms have been proposed to improve perceptual quality of scaled images at the expense of computational power.
Many of the content-adaptive downscalers employ non-adaptive algorithms like box filtering and Bicubic as fundamental tools. This means that the performance of those adaptive scaling methods can be improved by using a superior non-adaptive algorithm.
Kim Su Hyon, a researcher at the Faculty of Information Science and Technology, has proposed a new non-adaptive spatial filter kernel based on a circular area pixel model to improve the underlying frameworks of many state-of-the-art downscalers.
A pixel in a digital camera is the basic element of a sensor and its shape is rectangular. However, a point of light in a scene is spread by an optical system and creates a blurred circular image onto the pixel. In the spatial domain, an optical PSF (Point Spread Function) describes the degree to which an optical system spreads a point of light. Though the PSF from a circular aperture can be expressed by a sombrero function, it is usually modelled with simpler expressions such as a uniform circular disk. Therefore, from the optical point of view, he used a circular area pixel model rather than a rectangular one.
Since his kernel is one-dimensional, the proposed algorithm has two steps: horizontal and vertical processing. For upscaling, an original and a target pixel are treated as circular regions. For downscaling, only target pixel is treated as an elliptical region.
Abundant objective comparisons showed that the proposed downscaling algorithm is the fastest and has the highest PSNR and SSIM values among the commonest non-adaptive image scaling algorithms. Visual comparisons also showed that his algorithm produces the clearest images without blurring and halo effects. His filter kernel can replace the existing spatial kernels of edge-adaptive image downscalers to improve their performance further.
For further information, you can refer to his paper “ A New Image Downscaling Algorithm based on a Circular Area Pixel Model” in “ ACM Conference Proceedings” (EI).
Jo Aug 24, 2023
Thae Il Gwang, a researcher at the Faculty of Mining Engineering, has built an experimental jaw crusher for analyzing the working properties of jaw crushers commonly used in stoping faces and concentrating mills.
It is composed of a jaw crusher, sensors and a control circuit.
The crusher consists of a body with a fixed jaw, a movable jaw, an eccentric shaft, a connecting rod, a drawbar and a spring. Some structural modification was made to the drawbar and eccentric shaft to install necessary units for experiments.
Sensors are installed on the connecting rod and the eccentric shaft, respectively. They are used for measuring compressive force on the connecting rod and angular displacement on the eccentric shaft.
The control circuit can be connected with a computer to control the jaw crusher and to collect and send measured data.
The data sent to the computer is analyzed by an application for working properties and the computer displays angular displacement, angular velocity and moment of the eccentric shaft, crushing force on the movable jaw, and power consumption for crushing.
This device is used for analyzing fracture mechanism of different ores or for improving working properties of jaw crushers.
Jo Aug 23, 2023
A research team led by Kang Song Il, a researcher at the Faculty of Mining Engineering, has developed a load cell capable of precisely measuring mechanical properties of coal faces in full consideration of the effect of eccentric load in order to explain ground pressure characteristics.
The load cell consists of an elastic body and a resistance wire sensor.
The elastic body consists of four beams whose two sides are fixed and each beam has two resistance wire sensors on both sides. A total of 16 resistance wire sensors form a bridge circuit.
As four resistance wire sensors are connected in series on each branch of the bridge circuit, different deformations generated from them do not affect the overall measurement results. Instead, only total resistance change generated on the four resistance wire sensors is related to the measurement results.
The load on the elastic body is represented as the change of the sensor as it is transferred to the resistance wire sensor. Resistance values are calculated from the relationship between the resistance and the change, and displayed as resistance values in the multi-channel measuring device.
The load cell is installed in the support of a coal face. Vertical load is measured by a ground pressure meter between a beam and a middle support, which is set up in the middle of a beam of the support, while side load is measured by the meter in a lagging, which is set up between a coal wall and a bridge of the support.
The ground pressure characteristics in an area can be figured out by the properties of a certain area obtained in a mathematical method from measured values and the properties of known areas.
This device is helpful to get a correct description of ground pressure characteristics found in all kinds of coal mines and faces in different geological conditions, and to measure precisely the mechanical properties of corresponding rock mass such as cohesion, angle of internal friction and Poisson coefficient, which makes it possible to establish suitable methods of supporting mines. It will lead to a significant reduction in the consumption of materials such as supports, and safe working conditions and stability of coal production.
Jo Aug 21, 2023
A research team led by Kim Haeng Sok, a researcher at the Faculty of Electrical Engineering, has developed a partial discharge measurement system for insulation condition estimation in electrical power equipment.
Measurement of partial discharges (PD) is an effective way to evaluate insulating material degradation in high voltage equipment.
Partial discharge promotes degradation of insulating material and extends discharge paths to destroy insulation.
The partial discharge measurement system consists of a pulse current sensor, a measurement device and a display device. Partial discharge signals are transferred to the measurement device via a pulse current sensor, and after noise detection and correction, they are described as the amount of discharge to evaluate the state of insulation. A ferrite core is used for the pulse current sensor and the measurement device is aimed at filtering, amplifying and processing the signals.
The research team decided structural parameters of the pulse current sensor and detected the PD signals involved in noise by processing wavelet signals.
The system provides possibility to prevent hazards of high voltage equipment by evaluating in real time the state of insulation for any kind of electric power equipment. It can be widely adapted in the fields of electrical power industry.