Assessment, Characterization, and Improvement of Image Quality for Color/2.5D Printing and Scanned Document
In this dissertation, three related yet different problems are studied: Efficient approaches to scanned document healing, detection of ISO fade color pages, and characterization of BRDF parameters for 2.5D printing. In the first topic (the major one), we develop and optimize an image healing algorithm to fix defective scanned documents for hardware application based on exemplar image inpainting algorithm. To our best knowledge, the existing algorithms lack the feasibility in terms of time performance. In order to speed up the process, we use approximate nearest neighbor(ANN) search algorithm — multiple random KD-trees to partition the search space, so that we save time from irrelevant searches. The second topic is to evaluate printing image quality, on the specific aspect of ISO fading. A psychophysical experiment is conducted to get human perception on fading phenomenon, the result is used as ground truth. Then, in order to automatically detect faded images, a Machine Learning model based on SVM is trained with the ground truth. The main features are extracted in the projected color space, with the help of iterative K-means algorithm. The last topic is to study the key quality feature of 2.5D/3D prints: reflectance. This feature can be best characterized with Bi-directional Reflection Distribution Function (BRDF). We designed and printed two groups of relief printed test samples. We describe how BRDF data was acquired with our device for these samples. The correlation of surface parameters and BRDF data is discussed. These results provide a useful insight to the reflection and texture properties of relief prints, and can be further embedded in the printing pipeline.
Allebach, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our