Jiyoung Yoon

Medical Image Segmentation by More Sensitive Adaptive Thresholding

To make 3D surface model from medical images, segmentation is essential process. And, in the case that the HU(Hounsfield Unit) values of the object to be segmented varies in the image, adaptive thresholding segmentation is very efficient segmentation method. But in many cases, it shows insufficient sensitivity of segmentation that weak(dim) objects are not segmented. This problem is caused mainly by strong(bright) object nearby the weak object because adaptive thresholding makes high threshold due to strong object. In this study we present a more sensitive adaptive thresholding method overcoming this problem.

Development of image error correction system for 3D optical microscope

Recently 3D optical microscope is developing to see micro details in three-dimensions. To see images of an optical microscope in three-dimensions, we mount two cameras on the optical microscope. Three-dimension images are created using by two images that are incoming image through the refractive mirror with eye piece lens and those are made by two cameras. In this time, because of optical design errors and deformation mechanism of an optical microscope, it makes the problem which is disconformity of images. This case cannot be made a three-dimensional image correctly. We have corrected the error-output image of 3D optical microscope and developed the system showing the correct three-dimensional image.

Implement of Stereo Image Correction for 3D Optical Microscope and Design Hardware System

Recently 3D scenes are used in various industry fields such as medical, game, surface examination, biology and etc. 3D optical microscope can show very tiny details in 3D and it is developed variously. To look 3D images of an optical microscope, two cameras are mounted on an optical microscope. Incoming images through an object lens of an optical microscope are projected on sensors of mounted cameras by using refractive mirrors. Two cameras create left and right of each image and make 3D images. At this time, it can be a problem because of the optical design errors and deformation mechanism of an optical microscope. Then we cannot look 3D images. In this paper, we corrected this error with SURF algorithm and designed the hardware system to correct wrong mirrors position using servo motors.

Accelerating OpenVG with multimedia processors on mobile phones

Various handheld devices, in the first place mobile phones, have strong needs regarding 2D vector & bitmap graphics capabilities, including high performance and quality requirements. OpenVG is one of widely used low-level 2D vector graphics APIs. Currently, fully dedicated OpenVG semiconductor chips are relatively expensive and require high power consumption. In contrast, full software implementations show lower performance even with almost 100% of CPU usage, which would disrupt other concurrent applications. In this paper, we present a new cost-effective way of accelerating OpenVG, based on wide-spread and inexpensive multimedia-processing hardwares, presently on the mobile phones. Through an extensive use of these multimedia processors, we successfully accelerated our AlexVG, an OpenVG software-based implementation, especially on its fundamental OpenVG features: bit-block transfer, masking, scissoring, color conversion, image transformation, etc. This accelerated implementation comes with a lower power consumption. It exhibits appropriate performance, reaching more than 20 frames per second, even for complicated graphical user experiences. Its CPU utilization ranges from 20% to 30%, while the remaining CPU power remains available for other real-time tasks and user applications. This implementation is now commercially available and used in several mid-tier mobile phones.

Implementing a Stereo Image Processing for Medical 3D Microscopes with Wireless HMD

Recently 3D scenes are used in medical fields. We developed a medical 3D microscope with a HMD. We mounted two cameras on the 3D microscope. Incoming images through an object lens of an optical microscope are projected on sensors of mounted cameras by using reflector mirrors. These are processed in a computer and sent to HMD by wireless communications. However, there is the problem about 3D imaging display. Two camera images are easily distorted due to the defect in the 3D microscope and the color of the images are also a little changed because of the color characteristics of the cameras. In this paper, we suggested the method to correct the geometric and color distortion of the stereo camera image and real time implementation of these methods using GPU. We used a wireless HMD for ease of use by doctors. This system would be helpful for doctors to operate more comfortably.

ARINC661 Graphics Rendering Based on OpenVG and Its Use Cases with Wireless Communications

ARINC 661 is a standard for the CDS (Cockpit Display Systems) defined by ARINC (Aeronautical Radio, Inc.) and used for application to communicate and display sensing data and information. ARINC 661 contains various two-dimensional GUI (Graphical User Interface) widget definitions. The widget set covers a variety of graphics elements including circles, arcs, crowns and others. Additionally, the symbol widgets provide rendering of user defined polygonal shapes which are typically described with triangle fans and triangle strips. ARINC 661 has various rendering features including text, image output, transformation of object and the halo effect which renders the outlines of graphics objects for highlighting. It is possible to use 3D graphics libraries like OpenGL and DirectX for implementation of graphics features. However those 3D graphics libraries are too heavy and over-powered to show 2D graphics primitives defined in ARINC 661. In this paper, we propose ARINC661 rendering based on OpenVG. OpenVG is a standard of 2D vector graphics API defined by the Khronos Group. OpenVG is designed for embedded system GUI rendering and its features are appropriate to implement ARINC661. Compare with OpenGL, OpenVG defined with fixed pipeline architecture without programmable shader. Therefore, OpenVG is cost effective to rendering and certification of software quality by reducing complexity of application source code. We also propose ARINC661 use cases for wireless application to communicate and display data from various sensor network, information gathered from unmanned aerial and land vehicles.

Colorimetric Matching for Medical Cameras

While medical equipment and IT technology advances, doctors make medical surgery using input images through cameras. They use an optical microscope or an endoscope equipped with cameras. These systems display expanded images on a monitor for the required surgical sites. It is important to display almost the same color as the actual view, in case of making surgery or observing it. If monitors display wrong color, doctors cannot accurately recognize the surgical site and have difficulty to make a surgery, and surgery observers cannot get accurate information of the surgical site. In this paper, we proposed the characterization method for medical cameras using with the ColorChecker that is commonly used to calibrate cameras and additional color values on actual surgical site. It makes the estimated coefficient through polynomial regression. It can show surgical scenes to users more precisely.

ARINC661 Graphics Rendering Based on OpenVG

ARINC 661 is a standard for the CDS (Cockpit Display Systems) defined by ARINC (Aeronautical Radio, Inc.). ARINC 661 contains various two-dimensional GUI (Graphical User Interface) definitions, which are displayed with its graphics primitives called Widgets. There is displayed almost graphics primitives with GP and Symbol Widget. The widget set covers a variety of graphics elements, including circles, arcs, crowns, and others. Additionally, the symbol widgets provide rendering of user defined polygonal shapes, which are typically described with triangle fans and triangle strips. There is Halo feature that is rendering outline for highlight a graphic object. And it has variety rendering effect that are text, image output, object transformation, etc. Although we can use fully dedicated 3D graphics libraries, for instance OpenGL and DirectX for those graphics output. However those 3D graphics libraries are too heavy and over-powered to show 2D graphics primitives of ARINC 661. In this paper, we proposed ARINC661 rendering based on OpenVG. OpenVG is the standard of 2D vector graphics defined by the Khronos Group. It is appropriate for graphics rendering of ARINC661 because it has the standard conformance test suits. OpenVG can easily develop software and hardware version in comparison with OpenGL, so OpenVG is efficient rendering speed and development cost.

Accelerating OpenVG and SVGTiny with multimedia hardware

OpenVG and SVGTiny are widely used 2D vector graphics middleware and required high performance to support high resolution and high refresh rate. Currently available vector graphics hardware are relatively expensive and require high power consumption. In contrast, full software implementations show lower performance even with almost 100% of CPU usage, which would disrupt other concurrent applications. In this paper, we present a new cost effective way of accelerating OpenVG and SVG Tiny, based on wide-spread multimedia-processing hardware on the media devices and mobile phones. Through the effective use of these multimedia processors, we successfully accelerated OpenVG and SVGTiny, even with lower power consumption and lower CPU usage.

High Speed Vector Graphics Rendering on OpenCL Hardware

Most of computer graphics application is targeted to output for human eye which request 30~200Hz refresh rate and 1K ~ 4K resolution. However, in case of industrial application like a printing of circuit board, it requires rendering capability with much more big resolution and high performance with high precision of calculation. In this case, frequently, rendering using general graphics API is not fit for the requirement. In our research, we present case study of high precision, high speed and robust rendering of printed circuit board for high speed laser equipment. We used parallel programming using OpenCL API.