Adaptive image enhancement of text images that contain touching or broken characters
Text images that contain touching or broken characters can significantly degrade the accuracy of optical character recognition (OCR) systems. This paper proposes an adaptive image restoration technique that can improve OCR accuracy by enhancing touching or broken character images. The technique begins by processing a distorted text image with an OCR system. Using the distorted text image and output information from the OCR system, an inverse model of the distortion that caused the touching or broken character problem is generated. After generating the inverse model, the unrecognized distorted characters are filtered by the inverse model and then processes by the OCR system. To demonstrate its feasibility, six distorted text images were processed using this technique. Four of the text images, two with touching characters and two with broken characters, were synthesized using mathematical distortion models. The remaining two distorted text images, one with touching characters and one with broken characters, were distorted using a photocopier. The performance of the adaptive image restoration technique was measured using pixel accuracy and OCR improvement. The examples demonstrate that this technique can improve both the pixel and OCR accuracy of text images containing touching or broken characters.
- Research Organization:
- Nevada Univ., Las Vegas, NV (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- FC08-90NV10872
- OSTI ID:
- 42491
- Report Number(s):
- DOE/NV/10872--T176; ON: DE95010173
- Country of Publication:
- United States
- Language:
- English
Similar Records
Prediction of OCR accuracy using simple image features
Information Science Research Institute: 1994 annual report