GPU-Acceleration Generation And Editing Of Stylized Digital Images

As a multidisciplinary field, non-photorealistic rendering involves many research fields, such as computer art, computer graphics, computer vision, digital image, video processing and visual cognition. It has been broadly used in computer animation, movies, industrial design, games and advertisements.. Digital painting generation and editing are important research topics in this field. These topics have attracted lots of academic at-tention since the end of the last century, and numerous remarkable achievements have been made. This dissertation reviews the progresses of non-photorealistic rendering and makes an in-depth discussion about some key techniques. After the analysis, we elab-orate our new algorithms, including how to apply the artistic guidelines from artists in portrait drawing and hidden images generation, style-aware image cloning and style-aware painting. The main results of the dissertation are listed as follows:● In order to emphasize facial features like artists, we propose a portrait drawing generation algorithm based on facial features and line integral convolution (LIC). Firstly, our method defines an initial trimap using feature points of the human face and segments it using the Grabcut algorithm. Then, we render the portrait drawing using line integral convolution according to the segmentation. To adapt the line in-tegral convolution algorithm for portrait drawing, our method utilizes multi-scale white noise pyramids and vector field pyramids and proposes an adaptive edge extraction algorithm based on the segmentation. Experimental results demonstrate that our algorithm mimics the strokes of portrait drawing and emphasizes the facial features better, improving the effects of portrait drawings.● Hidden images contain one or several concealed foregrounds. Some regions of foregrounds are selected as clues by artist to assist viewers’recognition. To learn the artists’ rule of choosing recognition clues, our proposed algorithm applies vi- sual attention model to select the characteristic regions of the foreground. In ad-dition, satisfactory hidden images need appropriate hidden positions. To this end, we propose a novel approach for assessing the levels of recognition difficulty of the hidden images. Our system can suggest the appropriate hidden positions by estimating the recognition difficulty. We also improve the performance of hidden images synthesis using the parallelled texture synthesis algorithm. During the hid-den image synthesis, the spatial influence is also taken into account to make the foreground harmonious with the local surroundings. We validate the effectiveness of our approach by performing two user studies, including the quality of the hidden images and the suggestion accuracy.● We present style-aware image cloning, a novel image editing approach for art-works, which allows users to seamlessly insert any photorealistic or artificial ob-jects into an artwork to create a new image that shares the same artistic style with the original artwork. To this end, a real-time image transfer algorithm is develope-d to stylize the cloned object according to a distance metric based on the artistic styles and semantic information. In addition, several interactive functions, such as layering, shadowing, semantic labeling, and direction field editing, are provided to enhance the harmonization of the composite image. Finally, we designed two user studies to verify the effectiveness of our method. Experiments demonstrate that our algorithm improves the performance and the quality of the results.● We present a style-aware painting algorithm which help users to select the colors for the strokes. In addition, to simulate the directional effects of the artworks, we extract the directions of the strokes, including single stroke and lapped strokes. We have shown the results of many styles and extensive experiments demonstrate the effectiveness of our method.