Adaptive Region Division Stereo Matching Algorithm

Fig. 2. Region division results. (a) Reference image; (b) range 0 division result; (c) range (0,15) division result; (d) range [15,+∞) division result

Download full size

Fig. 3. Census transform result. (a) Pixel to be encoded; (b) pixel comparison result; (c) Census encoding result

Download full size

Fig. 4. Census transform results. (A) Reference image; (b) real disparity map; (c) traditional Census transform result; (d) improved Census transform result

Download full size

Fig. 5. Gradient point and cross arm of reference image. (a) Gradient point and cross arm of left reference image; (b) gradient point and cross arm of right reference image

Download full size

Fig. 6. Gradient calculation results. (a) Reference image; (b) calculation result of improved gradient; (c) calculation result of ABiGrad; (d) calculation result of traditional gradient

Download full size

Fig. 7. Cross area

Download full size

Fig. 8. Cost images. (a) Reference image; (b) cost image when d=6; (c) cost image when d=32; (d) cost image when d=58

Download full size

Fig. 9. Guided filter optimization results. (a) Reference image; (b) real disparity map; (c) before optimization of guide image filtering; (d) guide image filtering optimized

Download full size

Fig. 10. Experiment results. (a) Reference images; (b) real disparity maps; (c) results of proposed algorithm; (d) mismatching maps

Download full size

Table 1. Experimental parameter setting
View table
Table 1. Experimental parameter setting
Parameter $L_{1}$ $L_{2}$ $λ_{C e n s u s}$ $λ_{g r a d i e n t}$ $λ_{A D}$ $G_{T}$ $ε$ $τ_{S}$ $τ_{H}$
Value 17 34 30 1 9 10 0.01 20 0.4

Table 2. Comparison of Census algorithms
View table
Table 2. Comparison of Census algorithms
Algorithm Teddy Cone Venus Tsukuba Average
Traditional Census 12.14 14.22 1.10 3.55 7.75
Improved Census 11.22 14.24 0.98 2.96 7.35
Improved rate 0.92 -0.02 0.12 0.59 0.40

Table 3. Comparison of gradient algorithms
View table
Table 3. Comparison of gradient algorithms
Algorithm Teddy Cone Venus Tsukuba Average
Traditional Grad 14.86 14.10 2.72 3.42 8.78
ABiGrad 11.43 13.48 1.59 2.58 7.27
Improved Grad 11.80 13.22 2.03 2.48 7.38
Improved rate 3.06 0.88 0.69 0.94 1.4

Table 4. Comparison of optimization algorithms
View table
Table 4. Comparison of optimization algorithms
Algorithm Teddy Cone Venus Tsukuba Average
Scanline optimization 9.37 12.12 0.70 2.21 6.1
Guided filter optimization 9.35 11.91 0.61 2.06 5.99
Improved rate 0.02 0.19 0.08 0.15 0.11

Table 5. Mismatch rate of different algorithms on standard stereo picture pairs

View table

Table 5. Mismatch rate of different algorithms on standard stereo picture pairs

Algorithm

Tsukuba

Venus

Teddy

Cone

Average

Nonocc

All

Nonocc

All

Nonocc

All

Nonocc

All

RTCensus

Semi Glob

SO+borders

AdpDP

VariableCross

Adapt Weight

Assw-Grad

Len-ABiGrad

Proposed algorithm

5.08₉

3.26₈

1.29₁

2.05₆

1.99₅

1.38₂

1.57₃

2.12₇

1.66₄

6.25₉

3.96₇

1.71₁

4.22₈

2.65₆

1.85₂

2.00₃

2.50₅

2.06₄

1.58₇

1.00₆

0.25₁

1.92₈

0.62₃

0.71₄

0.89₅

0.25₁

0.26₂

2.42₈

1.57₇

0.53₁

2.98₉

0.96₄

1.19₆

1.00₅

0.62₃

0.61₂

7.96₈

6.02₃

7.02₄

7.32₆

9.75₉

7.88₇

7.20₅

4.97₂

4.47₁

13.80₅

12.20₃

12.40₄

14.40₆

15.10₇

18.30₈

12.40₄

11.00₂

9.35₁

4.10₆

3.06₂

3.68₄

6.41₈

6.28₇

3.97₅

3.68₄

2.78₁

3.35₃

9.54₃

9.75₅

9.18₂

13.70₉

12.70₈

9.70₄

9.18₂

8.68₁

11.91₇

6.34₈

5.10₅

4.51₃

6.62₉

6.26₇

5.62₆

4.74₄

4.12₁

4.21₂

Table 6. Running time of different algorithms on standard stereo picture pairs
View table
Table 6. Running time of different algorithms on standard stereo picture pairs
Method Teddy Cone Venus Tsukuba Enviroment
Adapt Weight 27.7 47.9 104.2 104.1 Interl i5 4 GB
Assw-Grad 4.2 3.9 2.8 1.7 Interl i5 8 GB
Len-ABiGrad 3.5 2.8 1.4 0.9 Interl i5 8 GB
Proposed method 3.1 3.1 1.2 1.5 Interl i7 8 GB