Application Example : Photo OCR
1. Photo OCR
1.1 Problem Description and Pipeline

Photo OCR pipeline
1.2 Sliding Windows

Pedestrian detection框的寬高比類似

Supervised Learning for pedestrain detection

Sliding window detection. 每次需要resize到輸入的固定大小

Text detection. Expansion

Character segmentation
1.3 Getting Lots of Data and Artificial Data

人工合成數(shù)據(jù)

通過變形原樣本擴(kuò)大數(shù)據(jù)集

加入噪聲

一般圖像加入噪聲不能改進(jìn)效果

獲取更多的數(shù)據(jù)
1.4 Ceiling Analysis : What part of the Pipeline to Work on Next (上限分析)

Ceiling Analysis

Another Ceiling Analysis example

Question.1
2. Summary

Main topics