As a first step in image processing applications it is often required to identify pixels above a threshold intensity level which contact one another. Sorting marker particles when the particle images cover many pixels is an obvious fluid mechanical application. With the present procedure, the image is scanned line by line once. The procedure is rapid, and can be continued across the boundary of the domain, for example, when the data is a long time series which must be artificially broken into "image" blocks for computer processing. The use of the algorithm is demonstrated on a variety of complex shapes, and estimates of speed of execution are given. The timing results show that the incremental time per bright region scales roughly as the square root of the number of bright regions. In addition, if bright regions fill more than (roughly) ten per cent of the total array space, the execution time per additional bright region becomes negligible. feel it is an improvement in several important respects. Each row is scanned once, and line segments of bright pixels are identified only by the end points of each segment. By keeping track of the bright segments of the previous row and comparing them to those of the current row, the method can properly group pixels belonging to each contiguous region. It is an inherently efficient computation, since segments from many bright regions are tracked simultaneously. Bright region connections are updated after each image row is processed, and no re-scan is required. As a result, conventional images can be double buffered for increased real time processing speed. Additionally, the image can be arbitrarily long in one dimension (the column dimension).