Dense disparity maps from sparse disparity measurements

Abstract

Abstract

Depth map upsampling using compressive sensing based model

“…M h ; D h in homogeneous and border regions are computed successively. Our sampling method is quite different from Hawe et al's work [11]. Their method relies on intensity edges for border detection, since no depth information is available.…”

“…This dictionary is exploited in a MRF model, which brings accuracy improvements for stereo depth estimation and range image denoising. Hawe et al [11] propose a CS-based depth estimation method from sparse measurements. They show that, by taking only 5% of the disparity data as measurements, their method can recover the full disparity map with high accuracy, which is quite impressive.…”

“…For Ψ and Φ, we follow the patterns defined in [11]: Ψ represents a Daubechies Wavelet basis, while Φ samples the high-resolution depth map with canonical pixel basis. The term Φ is important for our depth upsampling application as Φ should satisfy the minimum measurement requirement [24] deduced from CS theory.…”

“…Recently, some researchers have explored sparse representations for depth map processing [10,11]. Tošić et al [10] use sparse coding techniques [16] to learn a dictionary from Middlebury disparity data sets.…”

“…Our method is based on two key assumptions: first, neighboring pixels with similar colors are likely to have similar depth values; second, just like most natural images, an ideal depth map without noise corruption has large smooth regions and relatively few discontinuities, and therefore can be approximated with a sparse representation in some transform domain such as multiscale wavelets. Although the first assumption has been extensively explored in recent depth postprocessing work [4][5][6][7][8][9], relative less attention has been given to the second assumption [10,11].…”