ECSE-626 Literature Review Density Estimation Techniques Applied to Computer Vision Modeling

D ENSITY estimation is an established method in the fields of statistics and pattern recognition. Historically, the technique of histograms was used to convey the “general flavor of non-parametric theory and practice” [1, p47] of density estimation. A histogram precisely captures the essence of the density function by presenting the frequency of observations [1, p47]. With this idea in mind, we will draw upon techniques such as histograms, kernel density estimators, and newly developed estimators in statistics. For a treatment of histograms, refer to chapter 3 [1]. Moreover, for a discourse on kernel density estimators refer to chapter 6 [1]. Here, we will focus on the adaption of these classic non-parametric density estimation techniques to the world of computer vision. The underpinnings of histogram and kernel density estimators hold true to the literature [1], but in order to harness the power of these methods, they have been tweaked to fit problems of a different sort, ie. computer vision problems. The topic presented in this review is the application of density estimation techniques to computer vision modeling. Now that we’ve discussed some of the classic techniques along with introducing new approaches, we will show in this report, how such methods can be used within the field of computer vision. The driving question behind our investigation is the effectiveness of density estimation techniques in computer vision. We will explore this question through current and past research of density estimation techniques and through the discovery of trends regarding these techniques within computer vision. This allows us to gauge the effectiveness of such techniques in computer vision, and also to understand why they are used. The significance of posing our question is to further learn about the benefit and drawbacks of using the aforementioned density estimation techniques to model research problems in computer vision.