Robust Object Detection in Military Infrared Image

Robust Object Detection in Military Infrared Image

Tao Gao (Department of Automation, North China Electric Power University, Baoding, China)
DOI: 10.4018/japuc.2013010104

Abstract

The innovation of this paper is that we put forward a new algorithm of object detection form military infrared images with texture background according to the Mean-shift smooth and segmentation method combined with eight directions difference clustering. According to the texture characteristics of background, smoothing and clustering is carried out to extract the characteristics of object. The experimental results show that the algorithm is able to extract the object information form complex infrared texture background with better self-adapting and robustness. Future research particularly lies in raising the accuracy of object extracted.
Article Preview
Top

2. Mean-Shift Method

Given the japuc.2013010104.m01 samples japuc.2013010104.m02, japuc.2013010104.m03 =1, japuc.2013010104.m04, japuc.2013010104.m05, in the d-dimensional space Rd, then the Mean-shift vector at the japuc.2013010104.m06 is defined that:

japuc.2013010104.m07
(1)

japuc.2013010104.m08 is a district of high-dimensional spheroid with radius japuc.2013010104.m09, and satisfies the following relation of japuc.2013010104.m10 in order of gather:

japuc.2013010104.m11
(2)

japuc.2013010104.m12 is the number of points in the japuc.2013010104.m13 samples japuc.2013010104.m14, which falls into the japuc.2013010104.m15 district. (japuc.2013010104.m16 - japuc.2013010104.m17) is the deviation vector of the sample japuc.2013010104.m18. The Mean Shift vector japuc.2013010104.m19 which is defined in Equation (1) is the average value of the deviation vectors of the japuc.2013010104.m20 samples which fall into the japuc.2013010104.m21 district opposite the japuc.2013010104.m22 point. If the sample japuc.2013010104.m23 is got from the sample of the probability density function japuc.2013010104.m24, non-zero probability density gradient points to the direction at which the probability density increases most. On average, the samples in the japuc.2013010104.m25 district mostly fall at along the direction of probability density gradient, therefore the corresponding Mean-shift vector japuc.2013010104.m26 points to the direction of the probability density gradient (Comaniciu & Meer, 2002). From Equation (1), as long as it is the sample which falls into japuc.2013010104.m27 district, the contribution to japuc.2013010104.m28 calculation is all similar no matter its distance from japuc.2013010104.m29, however, the sample which is near the japuc.2013010104.m30 is more effective to the statistics characteristic that estimates japuc.2013010104.m31 surroundings, therefore the concept of kernel function is introduced. As important values of all samples japuc.2013010104.m32 are different, therefore a weighted coefficient is used for each sample. So the expand basic form of Mean-shift is:

japuc.2013010104.m33
(3) where:

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 11: 4 Issues (2019)
Volume 10: 4 Issues (2018)
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing