MOBILE PAGE SCANNER

MOBILE PAGE SCANNER. MOHAN PRASANTH D -181CS202 NANTHAKUMAR S-181CS216 RAGHU RAMKUMAR A K-181CS246.

Aim and Objectives. This project would be implemented on the Android platform, and the finished application would do the following while guiding the user through scanning a document page: Scanning of images. To sample photos, choose the appropriate sampling frequency/zoom level. To align your images, use transformation techniques (rotation, translation, and perspective correction). This accounts for the phone's movement (rotation/tilt) while taking the picture. Combine the processed photos to create a single, seamless image of the entire page. Apply OCR and other image-processing algorithms to the final product. This is an example of a potential extension to the OCR on a scanned page to transform handwritten documents is the goal of this research..

Algorithm. Input: n unordered images 1. Extract ORB features from all n images 2. Match the features between each pair of images and estimate pairwise Homography matrices using RANSAC. 3. Use bundle adjustment to find geometrically consistent Homography for each image. 4. Warp each image to global coordinate system. 5. Compensate for exposure difference to achieve uniform intensity. 6. Perform multi-band blending on all the images to smoothen all the edges to get the final stitched image. Output: Stitched image.

Scope of the project. Our team wants to create a mobile application that will assist users in scanning pages of documents as part of this project. The goal is to make the application's output as high-quality as that of a specialised scanner. Relevant image processing techniques would be employed to process the captured portions of a page of the document sequentially and concurrently, and the portions would be combined to generate a whole document with good resolution and clarity. Perspective correction, uneven lighting situations, text alignment, and other issues are expected to be addressed. This software would also provide a ready-to-use mobile framework for adapting various document processing algorithms to scanned documents..

Image stitching pipeline. The entire process of receiving close pictures of bits of a page from the user to generating a stitched whole page is referred to as the image stitching pipeline. In their study [2], Brown and Lowe describe this pipeline in great detail. In our project, we followed his instructions and tweaked a few steps to meet the needs of a mobile app. The following is a breakdown of each stage of the image stitching pipeline: Feature Detection, Image Matching and pairwise homography matrix computation, Bundle adjustment, Image wrapping..

User interface. Show* images f: Citk on Cvr•ra' to Start User Interface Ckar' buttm wil clear image stEk. 2: Click he Capüe' to 'nap step 3: are capt.n•g images, to stitchng Resutng be mce is Fig. 3. User Interface for the Android Application.

Post processing. The major methodologies involved in the post processing parts are: Gain Compensation: The intensity of overlapping areas between two photos should be the same. In fact,however, this is not the case. When aligned together, an intensity change that is not visible when capturing can provide a very noticeable boundary effect. Adjusting the intensity of each image to be roughly similar is a simple notion. We're attempting to minimize the error mathematically. Multi-Band Blending: Image edges are still apparent after gain adjustment.The intensity of boundary pixels is unlikely to be the same.We can blend all of the photos together to smooth out the edges. Blending is the process of assigning weights to various images on an overlapping area. Determining 1 to the centre pixel and varying the weights to zero at the boundary would be a basic method to setting weights for a specific image..

The Mobile Application. Here's a quick rundown of how to utilise the current version of the application, which is built on the Android platform. To activate/turn on the camera on the mobile device, the user first clicks on the camera button. The user then takes close grabs of the page by aiming the camera at specific areas and pressing the capture button. Each image acquired is added to the image stack, which updates the stack's visible length. Selecting the gallery button to inspect the recorded photographs and check that all areas of the document were captured with minimal noise/blurriness is an optional step. There is also an optional clear button that may be used to clear the stack of all collected photographs and restart the process. Finally, the stitch button is responsible for sending the collected images via the image-stitching pipeline..

Conclusion. The major goal of the study, which was to take high-resolution photographs of text documents using a mobile device, was definitely met.The stitched output image had far higher resolution than the whole image of the document taken at once, thanks to sophisticated image processing algorithms. In fact, the results are comparable to those of specialized scanner..

References. [1] R. Szeliski, “Image alignment and stitching: A tutorial,” Found. Trends. Comput. Graph. Vis., vol. 2, pp. 1–104, Jan. 2006. [2] M. Brown and D. G. Lowe, “Automatic panoramic image stitching using invariant features,” Int. J. Comput. Vision, vol. 74, pp. 59–73, Aug. 2007. [3] H. D. Jayant Kumar, Raja Bala and P. Emmett, “Mobile video capture of multi-page documents,” 2014..