CN104392448A

CN104392448A - Stereo matching method based on Gauss median segmentation guided filtering (GMSGF)

Info

Publication number: CN104392448A
Application number: CN201410704936.2A
Authority: CN
Inventors: 刘怡光; 王晓峰
Original assignee: Sichuan University
Current assignee: Sichuan University
Priority date: 2014-12-01
Filing date: 2014-12-01
Publication date: 2015-03-04

Abstract

The invention relates to a stereo matching method in computer vision, provides a stereo matching method of Gauss median segmentation guided filtering to the disadvantages of the stereo matching method of guided filtering at low-texture and discontinuous boundaries and the insufficient description of detailed information of median segmentation guided filtering and belongs to the field of computer application. In the method, firstly, BT (Birchfield and Tomasi) measuring theories are combined with two directional gradients (horizontal direction and vertical direction) to reduce noise influence and improve matching precision; secondly, a Gaussian weight window is introduced to the guided filtering; thirdly, superfine segmentation is introduced to replace fine segmentation in order to reduce hard segmentation limitation in the image segmentation; and finally, the advantages of the image segmentation and the median filtering are integrated and the Gauss median segmentation guided filtering (GMSGF) is provided. The experiment shows that the GMSGF has good stereo matching precision and detail preservation.

Description

The solid matching method of filtering (SMSGF) is guided based on the segmentation of Gauss's intermediate value

Technical field

The present invention relates to computer vision and Stereo Matching Technology, particularly relate to a kind of solid matching method guiding filtering (SMSGF) based on the segmentation of Gauss's intermediate value, belong to computer application field.

Background technology

Computer vision is the description obtained from the image gathered or image sequence three-dimensional world, and Stereo matching is the gordian technique of computer vision field research.Its target is in different viewpoint visual angles and different time places, in a width or a few width picture, find out corresponding point, and then obtains a dense degree of depth or parallax mapping graph.At present, robot vision, Autonomous Vehicles navigation, three-dimensional measurement, the field such as modeling and drafting based on image has been widely used in.

Solid matching method common at present comprises partial approach and global approach.Global approach mainly comprises figure and cuts (GC) and belief propagation (BP) method.But global approach precision is higher, but speed is relatively slow, is difficult to meet apply in real time, accurately and rapidly, especially in the field that three-dimensional reconstruction, three-dimensional video-frequency, robot navigation etc. are higher to real-time.Local algorithm mainly comprises bilateral filtering (BF), guides filtering (GF) etc.Wherein guide filtering to be one of part filter that the fastest effect is best at present, there is good edge effect.

Summary of the invention

The object of the present invention is to provide and a kind ofly guide the solid matching method of filtering based on the segmentation of Gauss's intermediate value, the solid matching method being intended to solve existing guiding filtering is in the problem of the deficiency of low texture and noncoherent boundary.

The present invention is achieved in that a kind of accurate fast solid matching method splitting guiding filtering (SMSGF) based on Gauss, and concrete steps are as follows:

Step 1: for improving precision and stability, we introduce BT (Birchfield and Tomasi) and estimate and two direction gradients (horizontal direction and vertical direction);

First, what guide the parallax effect of filtering to estimate cost has very high dependence, and on the basis guiding filtering cost to estimate, We conducted improvement, concrete function is as follows:

，

Then, the gradient that we improve cost function is estimated, and we adopt two Fextures (horizontal direction and vertical direction):

；

In color is estimated, we introduce BT and estimate:

Wherein, , be with the linear interpolation of its left direction point of proximity; be with the linear interpolation of its right direction point of proximity.

Finally, the cost of improvement estimate for:

.

Step 2: guide filtering to obtain parallax by Gauss and export:

Wherein ;

Utilize winner take all algorithm to obtain final parallax to be:

.

Step 3: ultra-fine segmentation is introduced into alternative meticulous segmentation;

In usual Image, meticulous segmentation obtains good effect, is widely used.But we find, meticulous segmentation still also exists larger probability and crosses over noncoherent boundary and region.So we have employed ultra-fine segmentation, make block little as far as possible, even if , with little as far as possible;

Step 4: the segmentation of Gauss's intermediate value guides filtering (SMSGF)

Usually simple medium filtering has preferably fast, robust, effectively to separate.Still in each cutting plate, get the intermediate value of parallax

Wherein, in segmentation in point parallax output valve.

The accurate fast solid matching method of the guiding filtering (SMSGF) based on the segmentation of Gauss's intermediate value that the present invention proposes, has good Stereo matching speed and precision, especially has good precision to detailed information.

Claims

1. the segmentation of Gauss's intermediate value guides filtering solid matching method, it is characterized in that comprising the following steps:

Step 1: for improving navigational figure precision, in conjunction with BT(Birchfield and Tomasi) estimate and two direction gradients (horizontal gradient and VG (vertical gradient)), color in estimating cost is estimated and gradient is estimated and improved, cost is estimated to estimate to estimate with gradient in color and is improved, to improve the precision guiding filtering simultaneously;

Step 2: for guiding filtering, because use box window, the parallax effect of acquisition is slightly partial to x/y direction. still adopt Gauss's weight window to carry out filtering;

Step 3: we guide on the basis of filtering in Gauss, introduce Iamge Segmentation (mean-shift algorithm), therefore ultra-fine segmentation is introduced into alternative meticulous segmentation; We find, in the ultra-fine segmentation obtained, in segmentation, to implement medium filtering afterwards, obtain the segmentation of Gauss's intermediate value and guide filtering (SMSGF).

2. Gauss's intermediate value according to claim 1 segmentation guides filtering, it is characterized in that, in step 1, the gradient in estimating guiding the cost of filtering is estimated and color is estimated and improved respectively:

In color is estimated, we introduce BT and estimate, and change cost and estimate, and concrete function is as follows:

，

Afterwards, we adopt two direction gradients (horizontal direction and vertical direction) to improve:

。

3. Gauss's intermediate value segmentation according to claim 1 guides filtering, it is characterized in that, in step 2, improve guiding filtering, meticulous segmentation still has greater probability to cross over discontinuity zone, therefore proposes ultra-fine segmentation, and carry out the good parallax effect of test acquisition, afterwards, we adopt Gauss's weight window to carry out filtering, obtain Gauss and guide filtering.

4. Gauss's intermediate value segmentation according to claim 1 guides filtering, it is characterized in that, in step 3, improves guiding filtering based on Iamge Segmentation:

In segmentation, implement medium filtering to parallax value, medium filtering value is as the parallax value of segmentation plane.