1 .

Introduction

1) cung cp mt phn loi cc thut ton hin c m thanh stereo cho php bc tch v so snh quyt nh thnh phn thit k thut ton c nhn 2) cung cp mt nn th nghim cho vic nh gi nh lng ca cc thut ton m thanh stereo. Hng ti mc ch ny, chng ti ang t trin khai mu ca cc thut ton tng ng cng vi d liu th nghim v kt qu trn theWeb www.middlebury.edu / stereo

2 Motivation and scope

Computational theory Bt k thut ton tm nhn, mt cch r rng hoc ngm, lm cho cc gi nh v th gii vt cht v hnh nh hnh thnh qu trnh, ni cch khc, n c mt l thuyt tnh ton c bnV d, lm th no cc thut ton o bng chng rng im ph hp vi hnh nh hai, tc l, h d on ca cc im cng mt cnh? mt ph bin gi nh l cc b mt Lambertian, tc l, b mt c xut hin khng thay i theo viewpoint. Mt s thut ton cng m hnh cc loi c th ca ting n my nh, hoc s khc bit v tng hoc ng cho 2.2 Representation Hu ht cc phng php th stereo tnh ton s chnh lch univalued chc nng d (x, y) i vi mt hnh nh tham chiu, m c th l mt trong nhng hnh nh u vo, Central to such methods is the concept of a disparity space (x, y, d) The term disparity was rst introduced in the human vision literature to describe the difference in location of corresponding features seen by the left and right eyes In computer vision, disparity is often treated as synonymous with inverse depth( Disparity c xem nh l cng khung nhn vi sau) More recently, several researchers have dened disparity(chnh lch) as a three-dimensional projective transformation (collineation or homography) of 3-D space (X, Y,Z) The enumeration of all possible matches in such a generalized disparity space can be easily achieved

with a plane sweep algorithm() The (x, y) coordinates of the disparity space are taken to be coincident with the pixel coordinates of a reference image chosen from our input data set. The correspondence between a pixel (x, y) in reference image r and a pixel (x,y) in matching image m is then given by (s ph hp gia pixel (x,y) trong nh r v pixel (x,y) trong nh m c cho bi x=x+sd(x,y) where s = 1 is a sign chosen so that disparities are always positive disparity space image or DSI(

The goal of a stereo correspondence algorithmis then to produce a univalued function in disparity space d(x, y) that best describes the shape of the surfaces in the scene(Mc tiu ca mt s tng ng m thanh stereo algorithmis sau sn xut mt chc nng univalued chnh lch khng gian d (x, y) m tt nht m t hnh dng ca b mt trong cnh.)

3 A taxonomy of stereo algorithm

Our taxonomy is based on the observation that stereo algorithms generally perform (subsets of) the following four steps 1. matching cost computation; 2. cost (support) aggregation; 3. disparity computation / optimization; and 4. disparity renement 3.1 Matching cost computation The most common pixel-based matching costs include squared intensity differences and absolute intensity differences(Ph bin nht da trn chi ph ph hp vi im nh bao gm s khc bit cng bnh phng v s khc bit cng tuyt i) these matching criteria are referred to as the mean-squared error (MSE) and mean absolute difference (MAD) measures; the term displaced frame difference is also often used(Trong cng ng x l video, cc tiu ch ny kt hp c gi l cc li c ngha l bnh phng (MSE) v c ngha l s khc bit tuyt i (MAD) Cc bin php, thi hn di di s khc bit khung cng thng c s dng)