Dense Stereo Matching Based on Cross-Scale Guided Image Filtering

Fig. 2. Initial disparity maps with different sizes of filter kernels. (a) l=r˙cmin; (b) l=r˙min; (c) l=r˙max; (d) initial disparity map with regularization term

Download full size

Fig. 3. Flowchart of proposed algorithm

Download full size

Fig. 4. Disparity maps of stereo matching algorithms based on image filtering. (a) Original color images; (b) real disparity maps; (c) disparity maps obtained by guided filtering cost aggregation; (d) disparity maps obtained by multiscale guided filtering cost aggregation; (e) disparity maps obtained by proposed algorithm

Download full size

Fig. 5. Comparison of running time and matching accuracy of several stereo matching algorithms based on guided image filtering

Download full size

Table 1. Percentage of mismatching pixels of different algorithms

View table

Table 1. Percentage of mismatching pixels of different algorithms

Algorithm	Tsukuba			Venus			Teddy			Cones			AvgPBM
Algorithm	n-occ	All	disc		n-occ	All	disc	n-occ	All	disc	n-occ	All	disc
Proposed	2.38	2.85	8.4	1.13	1.98	9.26	7.05	14.9	16.8	3.32	11.0	7.99	7.09
BF^[7]	2.93	3.37	9.40	1.89	2.67	9.43	9.39	17.2	19.6	5.29	13.7	12.3	8.94
GF^[10]	2.62	3.38	8.16	1.72	2.81	13.2	8.15	16.5	17.5	3.34	12.2	8.69	8.20
S-GF^[12]	2.30	2.94	8.12	1.09	1.99	9.67	7.04	15.1	16.7	2.99	11.4	8.17	7.29
GlobalGC $\begin{matrix} {P^{*}}^{[20]} \end{matrix}$	0.87	2.54	4.69	0.46	0.53	2.22	6.44	11.5	16.2	3.59	9.49	8.9	5.60
VarCros $\begin{matrix} {s^{*}}^{[21]} \end{matrix}$	1.99	2.65	6.77	0.62	0.96	3.20	9.75	15.1	18.2	6.28	12.7	12.9	7.60