During the image processing, in order to solve the problem of target automatic extraction and tracking under multiple scenarios, we present a method of image complexity description, in this paper. Based on the global features about the appearance of gray level, the appearance of target and the randomness of image texture, this method describe the image complexity using the information entropy, the edge entropy and the texture entropy. The experiments show that this method could measure image complexity qualitatively. The complexity given by this method is according with the difficulty of the work to extract and track the target.
Proc. SPIE. 7268, Smart Structures, Devices, and Systems IV
KEYWORDS: Image processing algorithms and systems, Digital signal processing, Detection and tracking algorithms, Image segmentation, Image processing, Medical imaging, Convolution, Target recognition, Automatic target recognition, Binary data
Real-time multi-objects labeling, which includes segmenting target, picking up features and labeling object
automatically, is a study field of Automatic Target Recognition (ATR). Labeling object in an image is a very important
step in many application areas such as target scouting and tracking, circuit board and IC mask inspection, environmental
and medical image analysis. In this paper, the approaches on object description are classified. Based on the comparison
of different methods in many aspects, some representative methods on object labeling, including their basic principles,
characteristics, and drawbacks are researched. At the same time, one optimized technique, fast object description on
multi-objects labeling, is presented. This fast object labeling algorithm is based on the pixel labeling method. Using the
neighboring relationship of binary pixel points in the contiguous scanning lines, this fast approach can complete the
pick-up of object labels and the combination of equal target labels during scanning once. Finally, in this approach, the
labeling of objects and pick-up of features are accomplished. Experimental results show that the improved algorithm can
be used to segment object and label multi-targets, and the performance of the algorithm is fast-speed, practical and
simple as well as with the high ability of processing complex patterns.