Correlation of portal images using fast fourier transform

The checking of the patient's position in relation to the treatment unit is mandatory in external radiotherapy quality assurance programmes. One way of checking positions is the electronic portal imaging. An automatic control could be performed simultaneously by the correlation of a reference 256 grey levels image with the current image of the treatment field of the patient. The algorithm used to process this correlation must be fast for a practical use (a few seconds). To accelerate the computation, the cross-correlation is performed by the means of the Fast Fourier Transform as applied by Jones and Boyer. As a consequence, the calculation time is divided by three. The principle of correlation of both images consists in comparing a significant squared area in the first image to a greater squared area in the second image by point to point scanning. In order to improve the convergence of this algorithm, we have added an optimization method. Because these methods have limitations depending on the starting point, we have chosen a compromise leading to an efficient and robust detection procedure. Instead of simple point to point scanning in the reference image, a quick scanning is proposed to detect the most probable fitting area. Then, the correlation function is optimized to find corresponding points between both images. Several points allow to evaluate the displacement in pixels, in relation to the reference image. Converted into distance, this displacement translates the deviation of the positioning of the patient