Skip to main content

Table 3 Classification rates for the test data set, using the hard-decision tree classifier of Fig. 1a [1]

From: Efficient and accurate document image classification algorithms for low-end copy pipelines

Ground truth

Classification rates, %

 
 

color-text

color-mix

color-picture

color-photo

mono-text

mono-mix

mono-picture

mono-photo

color-text

58/60

42/40

-/-

1/-

-/-

-/-

-/-

-/-

color-mix

-/1

98/98

2/1

-/-

-/-

-/-

-/-

-/-

color-picture

-/-

61/58

39/38

-/3

-/-

-/-

1/1

-/-

color-photo

-/-

42/36

-/-

58/64

-/-

-/-

-/-

-/-

mono-text

13/14

9/6

-/-

-/-

56/65

23/15

-/-

-/-

mono-mix

-/-

9/5

-/-

-/-

3/1

86/89

1/5

-/-

mono-picture

-/-

5/6

6/1

-/-

-/-

40/63

49/30

-/-

mono-photo

-/-

4/5

-/-

2/1

-/-

42/26

-/2

58/66

  1. Each entry in the table is “A/B” where A and B are the classification percentages, respectively, for the feature set proposed in the present paper and for the feature set obtained from [1]