Extracted Automatically Object To The Edge Of The Multi-level Domain Based On Bilateral Filtering And Its Description Of The Design Of The Operator

This thesis studies the digital image processing, focusing on image segmentation techniques and image representation and description. They are now widely used in computer vision, artificial intelligence, and multimedia information processing, etc. Based on the fundamentals of digital image processing, we propose a novel automatic object extraction algorithm, and a simple yet efficient shape descriptor.In this thesis, the object extraction algorithm focuses on extracting object boundaries from Canny edge field of an image automatically. A multi-scale approach is employed, where bilateral filtering is iteratively applied to the original image to create scales. These scales represent the process of successive simplification of an image. This means spatial details such as texture and noise are gradually removed, while major edges are maintained. In each scale, Canny edge detector is adopted to extract edges, which are then gathered in a scale map. By analyzing the edge field of the scale map, for example, the sub-segment formation and closed loop extraction, the object boundaries are finally captured.The shape descriptor developed in this thesis has the aim to represent the object boundaries computationally and reliably, and to describe the image semantically. The proposed approach is modeled similar to the process of a walking ant over the boundary of an object, where its limited line of sight at each time is cumulated in a two-dimensional (2-D) histogram.In order to ensure an accurate shape descriptor, two separate 2-D histograms are used to record the branches and corners of the object boundaries, respectively.Experimental results show that the proposed object extraction algorithm and shape descriptor are of high accuracy and reliability. They outperform their corresponding alternatives in literature.