text_segment_deeplearning_based

purpose

This code is written by tengshaofeng in order to segment character in a text line image.

reference paper

Chinese/English mixed Character Segmentation as Semantic Segmentation. Huabin Zheng

Requirements

Anaconda2.x tensorflow 0.x

prepare dataset

you can generate the dataset from several types of fonts by run: python dataset_gen.py the gen_imgs folder store the images, and corpus.txt store the characters. also you can manually label the 1024*48 text line image. dataset.txt save the label like： ['0.png', 0, 36, 45, 82, 87, 126, 129, 167, 172, 209, 215, 252, 253, 290, 305, 341, 354, 392, 393, 433, 435, 473] ['1.png', 11, 47, 56, 94, 99, 135, 141, 177, 179, 218, 232, 268, 273, 311, 326, 365, 368, 408, 421, 461, 461, 5] 0 , 36, 45 and so on is the split points(cols) in the image.

train

python CharacterSegmentTrain.py

experiment result