He then decided to try three radically different approaches. In the first, he used a pipelined solution: scaled the character to a nominal size, centered it in a grid, determined its axes of symmetry,...

He then decided to try three radically different approaches. In the first, he used a pipelined solution: scaled the character to a nominal size, centered it in a grid, determined its axes of symmetry, and rotated it to a standard orientation; then compared it to the set of “perfect” characters. In the second, he smoothed the printed character to eliminate noise from the suspect’s jittery printing by blurring it slightly using a mathematics-based filter operation, by applying still more mathematics to look at the character in a transform domain, and finally by comparing that to the transforms of the “perfect” characters. In the third approach, Eric decided to simplify things still further; he just counted the number of matches between cells on the 15 × 21 grid and the grid containing the printed character. He moved the printed character around over the “perfect” one until he got the best match, recorded the number of matches, and then repeated with the next “perfect” character until all 26 had been compared; the best match must be the winner, he thought. Give a brief analysis of each of his approaches with respect to the one with the best prospects for parallel processing.