Image reconstruction from projections using a genetic algorithm

The authors propose a new method of image reconstruction from projections using a genetic algorithm (GA). First, they encoded a binary image to a string by aligning pixel values in row-order. They then performed the following GA procedure. First, they prepared an initial population of strings at random. To pay-off each individual string and to create new strings in the next generation, the authors performed three GA operations, i.e., reproduction, crossover, and mutation. They evaluated each image (string) by comparing its calculated projection data with the original projection data. This process was continued until some stopping criterion, such as the maximum number of generations, was satisfied. Simulation results showed that the optimal image obtained by GA agreed well with the original image. The GA method was time consuming, but strings of the initial population were prepared with the image reconstructed by the filtered backprojection method (FBP). The convergence of the method using the initial population by FBP was about five times as fast as that using the initial population selected at random. From the simulations, the authors were able to reconstruct an image from projection data using a genetic algorithm