Font Size: a A A

Research On Building Extraction Techniques In Complex Scene Images

Posted on:2011-09-16Degree:MasterType:Thesis
Country:ChinaCandidate:L R HanFull Text:PDF
GTID:2178330332463517Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
As one of the important component in the construction of "Digital City", as well as a typical application in object extraction and recognition, target building area of an image extraction attracts increasingly attention. Based on the significant building extraction in simple scene images, this thesis focuses on algorithms for building extraction in complex scene images, and supplies reasonable pre-processing algorithms for two kinds of different complex building images separately.Building region extraction in complex scene images, on the one hand, can be used to reduce the search scope for feature point detection, matching and identification, which improve the efficiency of building target identification and 3D reconstruction. On the other hand, it is used to set up texture database of buildings for further achieving texture mapping automatically. The designed algorithm is demanded to extract the building region of the image accurately and quickly. Therefore, the algorithm for building extraction in complex scene images designed in this paper can extract building region as accurately as possible on the basis of excluded background interference. It is well known that the building region extracted is allowed to be slightly larger than its own outline.The core algorithms in this paper are based on the straight line features of the building's edges. In image pre-processing phase, in the view of the feature of the "shaded" building image, this thesis adopts the image binarization based on edge features and binary-image horizontal segmentation algorithm which can keep the image edge features perfectly and remove the bottom interference information effectively. For the characteristics of the "submerged" building images, the large-scale morphological sieve is used to achieve regional integration in order to preserve the building outlines without influence. In the straight-lines extraction phase, by analyzing two types of straight-line extraction algorithm, the author adopts Hough Transform algorithm to extract straight-lines. Meanwhile, a straight-line segment optimization algorithm is proposed to solute the problem that the classical Hough Transform is easy to produce "false segments". In view of the obvious color difference between the building region and the background in the images, this thesis proposes a method which can distinguish the building straight-line segments and the background straight-line segments effectively by analyzing the hue histogram cluster in the HSV color space.In the design of the building area extraction algorithm, considering the complexity of buildings' roof, taking the obvious advantages of vertical contour edges and the collection of lines after the hue cluster analysis, it can be divided into the following steps to extract the area of the building. First, fix the left and right borders of the building; second, fix the top and bottom borders of the building according to the boundary features and the correlative geometric constraint; Last, extract the region of the building automatically.
Keywords/Search Tags:Complex scene image, Morphological sieves, Hough transform, Hue cluster analysis, Building region extraction
PDF Full Text Request
Related items