input: a binary image
step by step pipeline:
step1: connected componet analysis
step2: recognition proceeds, two-pass process
the first pass: 努力识别每一个word
the second pass : 重新识别之前识别的不是特别好的word
2.1 Chopping chop points
Candidate chop points: 使用词(比如 hello )的轮廓的折线逼近的凹点。
2.2 Associating Broken Character
结合2.1和2.2 ,总体思路是fully-chop-then-associate
step3: 处理标记为含糊不清的空格