KT-Tech Incorporated

KT-TECH'S IMAGE REGISTRATION TECHNOLOGY

In image processing systems, frequently a problem arises when images taken at different times, by different sensors, and/or from different viewing points need to be compared. In order to detect the differences between the images, they need to be aligned with one another. A similar problem occurs when searching for a template or prototype in another image. The proper alignment between the image and the template must be found in order to obtain the optimal match for the template.

These problems and other related variations can be addressed by methods that perform image registration. Registration is a fundamental task in image processing. All large systems which evaluate images require the registration of images, or a closely related operation as an intermediate step prior to image processing.

For performing registration, a transformation must be obtained to relate the points in one image to their corresponding points in the other image.

The need to register images has arisen in many practical problems in diverse fields. These problems can be categorized as (i) integrating the information taken from different sensors, (ii) inferring three-dimensional information from images in which either the camera or the objects in the scene have moved, (iii) finding changes in images taken at different times or under different conditions, (iv) model-based object recognition.

KT-Tech is developing innovative and robust mathematical algorithms for image compression and image registration applications, such as remote sensing, e.g. global change, and medical image analysis, e.g. mammography and telemedicine.

KT-Tech' registration technique utilizes its proprietary nonlinear wavelet compression algorithm to remove the noise and redundancy in the images to be registered. The algorithm allows KT-Tech to determine the optimal compression ratio and the optimal number of pixels for registration. Using very high compression ratios, such as 1000-2000:1, and retaining very few hot pixels, such as 70-200 pixels, that contain the most significant information on the image, KT-Tech's technique can successfully register images.

Figures 1, 2, 3, 4, and 5 illustrate registration of a remotely sensed image (Figure 1) and its 18 degrees rotated version (Figure 2). KT-Tech nonlinearly compressed these images by a ratio of 1000:1, and then retained the 77 hottest pixels that contain the most significant information on the original image (Figure 3) and its rotated version (Figure 4). Finally, a statistical matching algorithm was utilized to compute the angle of rotation. As can be seen in the angle histogram (Figure 5), the peak is centered at 18 degrees, indicating the correct angle of rotation.

The applications of KT-Tech's technology include but are not restricted to (i) remote sensing data processing - autonomous aircraft landing, civilian and military applications, agriculture, geology, oceanography, oil and mineral exploration, pollution and urban studies, forestry and deforestation, and target recognition and identification; (ii) medical image analysis and diagnostic medical imaging such as tumor detection (for example: mammography) and disease localization, as well as biomedical research including classification of microscopic images of blood cells, cervical smears, and chromosomes; (iii) computer vision and pattern recognition for various different tasks such as segmentation, object and shape recognition, shape reconstruction, motion tracking, stereo mapping, and character recognition. In addition to these three major fields, KT-Tech's technology can be employed in speech understanding, robotics and automated inspection, computer aided design and manufacturing (CAD/CAM), and astronomy.