Efficient Optical Character Recognition (OCR) in images grabbed from Set-Top Boxes (STBs) plays an important role in STB testing. However, running OCR software on such images usually ends with low OCR performance since images can have low resolution, low image quality or colorful background. In order to improve OCR performance, four different image preprocessing methods are proposed. In this paper OCR is performed with Tesseract 3.5 and the relatively new Tesseract 4.0 on the images grabbed from different STBs. On the original images Tesseract 3.5 provides a 35.7% accuracy while Tesseract 4.0 attains a 70.2% accuracy. The proposed preprocessing methods improve OCR performance by 33.3% for Tesseract 3.5 and 22.6% for Tesseract 4.0 on the available images.

Improving optical character recognition performance for low quality images

In this paper, a preprocessing method is presented for improving Tesseract Optical Character Recognition (OCR) performance on images with colorful background. The proposed method consists of two steps. At first, a text segmentation method is performed which attempts to extract the text from the colorful background. This step is based on input image clustering into k images. In the second step, a classifier is used to identify the image containing text among k images resulting from the previous step. OCR is then performed on the identified image. The proposed preprocessing method improves Tesseract OCR performance by approximately 20%.

Optical Character Recognition on images with colorful background

In this paper, we present a novel approach for detection of moving vehicles in traffic videos. We propose a feature-based (corner-based) tracking to track and classify moving vehicles from the extracted ghost or cast shadow. The corner points of the vehicles are detected, labeled and grouped to generate a unique label per vehicle. This approach is able to deal with different types of deformations on the shape of the vehicles due to changes in size, direction and viewpoint. Also, the proposed method is totally free from motion estimation. To demonstrate the robustness and accuracy of our system, the results of the experiments are conducted on traffic videos including different complex background, illumination, motion, camera position, clutter and direction of the vehicles taken from outdoor boulevards and city roads. We detect moving vehicles on an average of 98.8% in a scene. The results show the robustness of our proposed algorithm.

Feature-based tracking approach for detection of moving vehicle in traffic videos

Existing texts play an important role in understanding a scene image. Scene images differ from document images, which are composed of text characters of various size, shape, direction and situation along with complicated backgrounds, such as map, picture or painting, etc. Hence, the extraction of texts in scene images is a difficult as well as challenging task. Recently, mathematical morphology based algorithm finds applications to extract texts from scene images. In this paper we worked on a modified morphological filter to improve extraction accuracy. Due to insufficiency of a single threshold value, we have divided input images into different clusters depending on the size of texts.

Extraction of texts from a scene Image using morphology based approach

Quite often it is observed that text information in documents is printed on colorful complex background. Smooth reading of text content in such documents is difficult due to background patterns and mix up of foreground text color with background color. Further the character recognition rate when such documents are OCRed, is low. In this paper we are presenting a novel approach for extraction of text information in complex color document images. The proposed approach is a three stage process. In the first stage the edge map is obtained utilizing the Canny edge operator. The edge map is split into blocks of uniform size and image blocks are classified as text or non-text. In each text block the possible text regions are identified and enclosed in tight bounding boxes using x-y cut on edge pixels. Further the text regions that are immediate adjacent to each other in vertical direction in which the character(s) are split horizontally are merged so as to enclose the character(s) fully in one text region. In the second stage certain amount of false text regions are eliminated based on a property of printed text. In the last stage the foreground text in each text region is extracted by unsupervised thresholding using the data of refined text regions. We conducted exhaustive experiments on documents having variety of background complexities with printed foreground text in any color, font and tilt. The experimental evaluations show that on an average 98.03% of text is identified. The processed document images showed better performance when OCRed compared with the corresponding unprocessed source document images.

/pdf/foreground-text-extraction-in-color-document-images-for-1ydgg41fkc.pdf

Foreground Text Extraction in Color Document Images for Enhanced Readability

Reading of the foreground text is difficult in documents having multi colored complex background. Automatic foreground text separation in such document images is very much essential for smooth reading of the document contents. In this paper we propose a hybrid approach which combines connected component analysis and an unsupervised thresholding for separation of text from the complex background. The proposed approach identifies the candidate text regions based on edge detection followed by a connected component analysis. Because of background complexity it is also possible that a non text region may be identified as a text region. To overcome this problem we extract texture features of connected components and analyze the feature values. Finally the threshold value for each detected text region is derived automatically from the data of corresponding image region to perform foreground separation. The proposed approach can handle document images with varying background of multiple colors. Also it can handle foreground text of any color, font and size. Experimental results show that the proposed algorithm detects on an average 97.8% of text regions in the source document. Readability of the extracted foreground text is illustrated through OCRing.

Nirmala Shivananda

Papers

Separation of Foreground Text from Complex Background in Color Document Images