Citation Context

...struction, we perform anisotropic diffusion of the symmetrized cost function Fy(x1, x2) ≡ C(x1, y, x2 − x1), aggregating costs along the ridge direction predicted by the normal estimates. As shown in =-=[20]-=-, given a pair of rectified cameras, the unit tangent direction t to the matching ridge at (x1, x2) can be expressed as t ∝ (n ⊤ p 1, n ⊤ p 2) ⊤ where n and p i are, respectively, the 2D projections o...

Citation Context

...g accurate sub-pixel disparities since the parabola fitting approaches exhibit artifacts known as pixel-locking [15]. These approaches perform a careful interpolation of the integer disparity results =-=[10]-=- or work on a disparity space image that is sampled at fractional disparities [17]. How is this paper organized? Section 2 gives an overview of stereo sub-pixel interpolation techniques. In Section 3,...

Citation Context

...nce 5 ger estimates and can create noticable ”stair steps” on surfaces that should be smooth [9], [10]. One way of attenuating the pixel-locking effect is through the use of a symmetric cost function =-=[11]-=- for matching the “left” and “right” image blocks. To avoid the high computational complexity of these methods another class of approaches based on the Lucas-Kanade algorithm [12] proposes an asymmetr...

Citation Context

...r integer estimates and can create noticeable ”stair steps” on surfaces that should be smooth [8], [10]. One way of attenuating the pixellocking effect is through the use of a symmetric cost function =-=[5]-=- for matching the “left” and “right” image blocks. To avoid the high computational complexity of these methods another class of 1 This work was funded by the NASA Lunar Advanced Science and Exploratio...

Citation Context

...s effect [35] performs a linear interpolation and maximization of an enhanced NCC similarity measure. A more complex solution to avoid pixel-locking is the symmetric refinement strategy considered by =-=[32]-=-. Indeed, the images are treated symmetrically and a 2D surface fitting of the function cost is required to get sub-pixel coordinates for both matched points. The sub-pixel accuracy is even more rare ...

Citation Context

...cise location of points, in the secondary image, that correspond to discrete positions in the reference image. In chapter 3, which is based on a collaboration with Szymon Rusinkiewicz and James Davis =-=[90]-=-, we show that this strategy can lead to a systematic bias associated with the violation of the general symmetry of matching cost functions. This bias produces random or coherent noise in the final re...

Citation Context

...ade to address this problem. For example, Shimizu and Okutomi [141] attempted to reduce the bias by performing additional curve fitting on matching costs defined at half-pixel locations. Nehab et al. =-=[142]-=- suggested symmetric refinement by fitting a parametric surface over a 2D neighborhood of the matching cost function. Stein et al. [143] proposed an iterative refinement method that is essentially bas...

Citation Context

... use of optical flow. These methods do not discuss whether high-resolution images help improve the flow estimate or not. Methods that use interpolation to enhance the accuracy in computing data costs =-=[3,23,25,18,19]-=- also do not discuss the effect of the regularization terms and still solve the final energy functions at the original resolution. In contrast, we show that performing matching with high-res images no...

Citation Context

...er area unit, such as pixels per inch or per square inch. None of these pixel resolutions are true resolutions, but they are widely referred to as such; they serve as upper bounds on image resolution =-=[2]-=-. SR is a technique that constructs high-resolution (HR) images from several observed LR images, thereby increasing the high frequency components and removing the degradations caused by the imaging pr...

Citation Context

...sary for processing real-world data sets such as the Apollo Metric Camera data. We investigated a large number of stereogrammmetric systems that can provide dense stereo matching from orbital imagery =-=[7, 8, 9, 10, 11, 12]-=-. A common technique in sub-pixel refinement is to fit a parabola to the correlation cost surface in the 8-connected neighborhood around the integer disparity estimate, and then use the parabola’s min...