DOCUMENT TEXT DETECTION IN VIDEO FRAMES ACQUIRED BY A SMARTPHONE BASED ON LINE SEGMENT DETECTOR AND DBSCAN CLUSTERING

Abstract

Automatic document text detection in video is an important task and a prerequisite for video retrieval, annotation, recognition, indexing and content analysis. In this paper, we present an effective and efficient model for detecting the page outlines within frames of video clip acquired by a Smartphone. The model consists of four stages: In the first stage, all line segments of each video frame are detected by LSD method. In the second stage, the line segments are grouped into clusters using the DBSCAN clustering algorithm, and then a prior knowledge is used in order to discover the cluster of page document from the background. In the third and fourth stages, a length and an angle filtering processes are performed respectively on the cluster of line segments. Finally a sorting operation is applied in order to detect the quadrilateral coordinates of the document page in the input video frame. The proposed model is evaluated on the ICDAR 2015 Smartphone Capture OCR dataset. Experimental results and comparative study show that our model can achieve encouraging and useful results and works efficiently even under different classes of documents.

Keywords

• Document text detection
• video text segmentation
• Smartphone
• line segment detector
• DBSCAN clustering