Image Filtering Improvement For Producing Artistic Images

The need for processing digital images has increased dramatically due to the growth of computer graphics and digital cameras. Among the existing image processing techniques, the technique to convert an input image to the artistic style image has gained much attention lately. Computer technology now allows the quick and easy creation of highly realistic images of natural and imaginary scenes. Plenty of research work has been proposed to create artistic styles with computers. Current work refers to many image processing techniques, which transform an image into a specific artistic style (or more precisely a depiction style), such as ink painting, pointillism, colorization, image composition, charcoal drawing, etching, etc., or abstractionism more generally, in a style other than realism. Ideally, a human user could choose from a wide range of visual styles, leaving image creation details to computers. Aided by computers, it is simple to colorize an old feature-length movie nowadays, which would be prohibitive with traditional media.Filtering is one of the most important operations in image and video processing. In particular, Edge-Preserving Smoothing Filters (EPSF), as a fundamental building block for several applications, has received considerable attention from the research community over the last two decades. The primary object of this dissertation is to study the relationship between image filtering and artistic images in terms of their similarities and differences theoretically and experimentally. We analyze the strengths and weaknesses of existing image filters and propose some new algorithms to produce artistic images. We also discuss the quality of our algorithms through evaluating the effectiveness of our techniques in producing artistic images, including image colorization, image abstraction, image segmentation, and video clips colorization. The main challenge is the tremendous effort that required during producing artistic images. For instance, a single image needs hours or even days of painstaking work by a photo-artist. To solve the problem, we proposed some new simple yet surprisingly effective interactive techniques that drastically reduce the amount of input from the user. In addition, our techniques are also extremely useful in digital photography. The main contributions of this dissertation are listed as follows.· Two techniques for interactively colorizing a grayscale image are proposed by introducing the Nikolaou nonlinear filter. In our two algorithms, instead of removing noise or unnecessary details from the images as proposed by existing work, we use the Nikolaou filter to automatically expand the color hints from the user's paint brush and align with the object boundary. To improve the efficiency of colorization, we first propose an algorithm to reduce the manual effort of colorization by automatically selecting desired color within a little computation time. Second, we propose a new algorithm to interactive colorization by introducing a new concept to the bilateral filter. We adopt the bilateral filter as a range weights filter (bilateral filter without Gaussian spatial domain). Compare with the Nikolaou filter, our algorithm can achieve better filter near the edge and the equivalent effect in performance. Our proposed algorithms can improve the quality of colorization results for still images as well as the computational cost.· A non-iterative filter is proposed for image abstraction through removing the details in low-contrast regions. An edges-preserving filter is developed to extract useful image contents, which guides the abstraction to embody the artist's intention. We propose a non-iterative filter that removes the details in low-contrast regions without filtering across discontinuities by adapting dynamic shape and metric space of the filter to the local structure of the input. The proposed filter generalizes the well known Symmetric-Nearest-Neighborhood-Filter (SNN) without suffering from the related drawbacks. Finally, we demonstrate the utility of our non-iterative filter in creating effective yet simple to implement solutions for common image processing tasks. To exercise our formulation we have created a new saliency-based sharpen filter and a pseudo image abstraction application. The behavior of our filter is not only suitable for preserving the edges, but also suitable for sharp the corners. Compare with the previous works, our approach could get a better abstraction of real images efficiently.· We propose an interactive method for image segmentation, which adopts a statistical method to separate image foreground. Based on "edge-stopping" functions, our method can automatically control the expanding of seed colors with the boundaries of the image. The proposed method is capable of removing unwanted complicated objects from the image and pulling out complicated features, such as texts, with low user efforts. Our proposed method could obtain excellent segmentation results at low computational cost. The results show the extra versatility and the ability to segment objects in highly complex situations.· For saving the time and effort in colorizing video clips, we propose a semi-automatic video colorization algorithm. Our method propagates the colors from key frame to the rest of the sequences of grayscale images, in which the same objects between adjacent frames move slightly. Therefore, when processing the posterior frame, we take the preceding one as the reference image and assign the chromatic information from reference image to unmoving objects directly. For moving objects, we compute the flow field and assign the corresponding chromatic information accordingly. Experiments on various videos demonstrate that our colorization method achieves much high-quality, yields more reliable colorization performance results than the conventional state-of-the-art methods.· We have demonstrated an interactive system by combining advanced computational algorithms with novel user interfaces for quickly producing artistic from images and videos. Our systems are capable of producing high quality results with a limited amount of user input such as a few paint strokes of the mouse. Our system provides interactive tools to produce artistic images in accuracy and efficiency.In conclusion, this dissertation focuses on developing the image filtering, in particular, Edge-Preserving Smoothing Filter (EPSF) for producing artistic images. Several efficient and effective solutions are proposed for the problems of colorization, image abstraction, and image segmentation. Theoretical analysis and experimental results prove the outstanding superiority of our proposals over other related works.