Automated Degradation Diagnosis in Character Recognition System Subject to Camera Vibration

Algorithm 1

Motion-blurring diagnosis procedure.

Input: an input character image 𝐼_𝑖

Output: motion-blurring level 𝐿_𝑖

Step  1. A given character image 𝐼_𝑖 is preprocessed to a uniform image 𝐼^𝑢_𝑖 (64×64 pixels).

    Four edge structure elements are used on 𝐼 to acquire the edge images 𝐸^−45∘_𝑖, 𝐸^0∘_𝑖, 𝐸^45∘_𝑖, 𝐸^90∘_𝑖.

Step  2. The edge density ratio feature 𝑟¹_𝑖 and 𝑟²_𝑖 can be computed by (i) and (ii):

   𝑟¹_𝑖=∑ℎ^0∘_𝑖∑ℎ^90∘_𝑖                    (i)

   𝑟²_𝑖=∑ℎ^−45∘_𝑖∑ℎ^45∘_𝑖                   (ii)

   Here, ℎ^−45∘_𝑖, ℎ^0∘_𝑖, ℎ^45∘_𝑖and ℎ^90∘_𝑖 are the respective gray distribution features of

   𝐸^−45∘_𝑖, 𝐸^0∘_𝑖, 𝐸^45∘_𝑖, 𝐸^90∘_𝑖, namely if the bin of

   𝐸^−45∘_𝑖 gray histogram is more than defined thresholding 𝑡_bin, it would be recorded 1.

Step  3. The one big level after disk-blurring diagnosis can be further diagnosed as followed:

     if 𝑟²_𝑖<𝑡₄,𝐿_𝑖=𝐿₄

     elseif 𝑟¹_𝑖<𝑡₅,𝐿_𝑖=𝐿₅

     elseif 𝑟²_𝑖>𝑡₆,𝐿_𝑖=𝐿₆

     elseif 𝑟¹_𝑖<𝑡₇,𝐿_𝑖=𝐿₇

     else 𝐿_𝑖=𝐿₂

   Here 𝑡₄, 𝑡₅, 𝑡₆ and 𝑡₇ are the thresholdings for motion-blurring diagnosis in four directions.