A Process to Remove Direct Hit Noise From Image Sequences

University of Arizona
posted on 05/24/2010

• request more info 
•   
•  LICENSE ONLINE

Background:  In certain types of x-ray detection and imaging, a spotting problem can arise from the incidence of x-rays on the detection medium. In certain types of detectors, namely those employing CCDs, CIDs, or other solid-state imaging devices, photons of radiation that directly impact the detecting medium can cause high levels of signal within that pixel, which produce artifacts in the radiological image, that is, bright spots that do not represent structure in the actual object being x-rayed. This limits the applicability of such detectors in a number of applications, both medical and industrial. The typical solid-state x-ray imaging detector array uses a two-step process to create the video image. First, the x-ray photons strike a conversion layer that converts the x-ray radiation to a visible light output, and then the visible light falls on a light sensing array, e.g., a CCD. A step can also be taken to keep the x-ray radiation from interacting directly with the light sensing array. For example, the conversion layer can be constructed to absorb the x-ray radiation, or alternatively there can be a layer of a radiosorptive material, e.g., leaded glass, interposed between the conversion layer and the light sensing array. This material is intended to pass light in the visible spectrum, but to absorb the x-ray photons. Ideally, the light sensing array is designed to react photoelectrically only to light in a specific wavelength band, e.g., only to green light (about 560 nm), with the x-ray photons passing through without generating any electrons or holes. However, in reality some portion of the x-ray photons that are incident on the sensor will create a signal on the light sensing array. Because the x-ray photon has a much higher energy than that of a visible light photon, the picture element or pixel involved will become flooded, i.e., will appear as a white spot at that location on the image.

Invention Summary:  Accordingly, it is an object of this invention to provide a technique to minimize or eliminate the spot noise described above and to avoid the drawbacks of current technologies.  The concept of this invention can be employed with standard scintillators, such as Gadolinium Oxysulfide, Cesium Iodide, Thallium-doped Cesium Iodide, or other available material. The photosensor array can be a standard device as well, such as a CCD or a CID. In a device of this type, an image is acquired by integrating a photoelectric signal over some integration period T. This period can be divided or segmented into N multiple successive frames each of a shorter period T/N. For example, the imaging array could integrate for a time T/4, then read out and store image #1, integrate for a second time T/4, read out and store image #2, integrate for time T/4, read out and store image #3, and then integrate for time T/4, read out and store image #4. The total integration time is still T. Because the images 1, 2, 3, and 4 are taken in quick succession, they can be easily acquired in registry with one another.

The imaging array is arranged to capture the image as a grid of picture elements or pixels. The grid is generally an array of n lines, with each line containing m pixels. Each pixel thus has a given pixel location (M,N) where N and M have any integral value from 0 to n-1 and 0 to m-1, respectively. The number of pixels depend on the desired resolution. For example, the grid may be an array of 512 by 512, with pixel locations from (0,0) to (511, 511).

It has been observed that the x-ray hits (which cause the spot noise described above) are random events. The x-ray hits may occur at any given pixel location in any given frame. Because of the randomness of this phenomenon, it is statistically unlikely that an x-ray hit will occur at the same pixel location in each of two successive frames. It is even more unlikely that an x-ray hit will occur at the same pixel location in N out of N frames, for a number N more than two.

Because the video signal is stored and recorded as a signal strength, or brightness value, for each of the (n×m) pixel locations, it is possible to examine the various image frames pixel by pixel, and examine each pixel in each of the several frames. The respective pixel values can be ranked from highest to lowest, in terms of brightness value. A rank order filter technique is employed to select a certain place in the ranking for the pixel values at each pixel location. For white spot noise elimination, the best selection is to pick the minimum value of the set of frames, and assemble a resulting image based on the selected lowest value pixel for all of the pixel locations. Because of the randomness of the white spot noise, and the order that is inherent in the image, the resulting output image will be substantially free of white spot noise. Even where there are a large number of white spots in the raw image, the white spots are greatly reduced where the filtering selects the lower ranking pixel in two frames (N=2), and are virtually eliminated where the filtering selects the lowest ranking pixels in three frames (N=3). This filtering eliminates the white spot noise without degradation and without loss of sharpness in the underlying image.

The image processing or filtering can be carried out in any convenient computer or digital signal processor. In one possible embodiment, the two, three or four (or more) images in registry are acquired and stored digitally. Then, the values from the first pixel for all four images are compared, and the minimum (or other selected rank) value is stored in the first pixel location of an output image. All remaining pixels are processed in like manner, thus yielding the output image. Another possible embodiment, which requires less storage and less latency, is to bring the first image into a frame buffer as it is acquired. Then, as the second image is acquired, the pixel values of the incoming second image are compared to the values of the stored (first) image. The minimum of the two values is placed back into the frame buffer at that pixel location. After the second image is complete, the frame buffer contains the minimum value (pixel-by-pixel) image of the two frames. As a third (and subsequent) image is acquired, the pixel values are compared pixel by pixel with the stored image. If the incoming pixel value is smaller, the smaller value is placed into the frame buffer at that pixel location. This sequence continues until the last of the N images is acquired. The frame store then holds the minimum (pixel by pixel) image over the acquired frames.

As mentioned before, the pixel value that is selected need not be that of lowest rank or value. For some applications, it may be desirable to select the pixel of highest rank, or of an intermediate rank. This technique can be used for eliminating or reducing blemishes or artifacts that occur for other reasons, and should not be considered limited only to the white spot noise problem discussed above.

IP Status:  Patent issued and available for licensing.

Refer to UA Case Number UA97-043