Digit recognition using fractal and moment invariants

Abstract EN

This paper presents quantitative and qualitative methods based on fractal geometry and invariants moment to recognize the printed or handwritten digits.

The fractal features are computed for the area of image segment within the frame.

The analysis of the results showed that the fractal dimension can be recognized different digits printed at same font size, but it cannot distinguish the printed digits or handwritten at different font sizes.

The lacunarity appears high ability in recognizing printed or handwritten digits at different font sizes.

The investigation of fractal description efficiency necessitates comparing the fractal performance with a common moment descriptor used in this field.

The comparison proved that the fractal geometry possess high digit recognition capabilities and it gave 93 % score of recognizing the printed digits during 6.6 s and 71 % to recognize handwritten digit during 6.8 s, which is greater than the corresponding values 81 % and 64 % scores during 1.5s and 1.7 s, respectively given by the moments.