The disclosure is directed to an improved method and apparatus for obtaining a processed image of a body. In accordance with the disclosed method, a bolus of contrast medium is injected into the body, and the body is irradiated. Radiation which passes through the body is detected and converted into an electronic video signal. The generated video signal is then filtered with a filter function having a temporal frequency response that has a single bandpass peak at a temporal frequency substantially corresponding to the temporal frequency of the movement of the bolus of contrast material through a region of the body. The filtered signal can then be displayed and/or recorded.

The method and the device in accordance with the present invention utilizes a recursive filter for forming a composite image from a short series of X-ray images with the weighting factor .alpha. used in the filter being a function of the number of X-ray images processed and to be processed. When .alpha. is suitably chosen, the amplitude of the composite image is always constant, such as a constant image brightness, and each X-ray image makes an equally large contribution to the composite image. The noise level in the composite image decreases as the number of processed X-ray images is larger.

Two image processing devices having different time delays are used to generate time difference images having high signal-to-noise ratios. The image processing devices may be real time image processors having only one image memory wherein consecutive images are sequentially fed in digital form to a recursive filter. In such devices the memory is employed in the recursive filter as a "delay element", as a result of which noise integration and differential image determination between consecutive images takes place.

A filtering method and a modified automatic regressive filter therefor are provided. The modified automatic regressive filter includes: a frame memory for storing one frame of the output of the filter; a frame difference generating portion for receiving the signal stored in the frame memory and an observed signal, to output a frame difference, and determining whether the current pixel belongs to a non-moving region or a moving region to output a moving index; a histogram generator for converting the frame difference of the non-moving region into a cumulative distribution function required for calculating the variance of noise in the non-moving region of the observed signal; and a selective filter output portion for selectively outputting the observed signal unaltered in the moving region and the signal passed via an adaptive filter having the gain calculated based on the output data of the histogram generator in the nonmoving region. Thus, a continuous moving signal is processed in real-time with high noise removing capability, and the edge of the moving target is preserved without being blurred.

An electro-optical system which integrates the image pattern information, extracted from a plurality of image frames and identified as belonging to a common candidate object, to form a composite image pattern of the candidate object which may be accurately classified with respect to a reference object image pattern is disclosed. More specifically, the formed composite image pattern of the candidate object is correlated with each image pattern of a set of reference object image patterns to generate a corresponding set of correlation values. Preferably, the largest of the correlation values is selected for a comparison with a threshold correlation value. In the event the generated correlation value exceeds the threshold correlation value, the candidate object represented by the composite image pattern is classified as the referenced object corresponding to the selected correlation value. Another aspect of the electro-optical system permits the threshold correlation value to be adaptively selected in accordance with the number of frames integrated to form the composite image pattern under consideration for classification. In addition, a final classification of the candidate object may include the smoothing of a plurality of classification decisions generated throughout the integration sequence during which a number of composite image patterns are formed.