pointfeaturematchingmethods1-文庫(kù)吧
2025-09-08 15:11 本頁(yè)面
【正文】 the L2 norm CVPR04 Stockman (27 June 04) 10 How to determine landmarks ? By unique intensity/color structure of neighborhood ? By geometry or topology of neighborhood (part 2 below) Image 1 Image 2 Vectors show how intensity neighborhoods moved: similar vectors are colored the same Do these two images meet the definition of “registration” CVPR04 Stockman (27 June 04) 11 (1) Determine salient points (2) Match neighborhoods CVPR04 Stockman (27 June 04) 12 Vector mathematics Sc hw ar t zC au c hybyandbe t w e e nisw hi c hTSTSnc or r e l at i oc r os sno r m al i z e dSTdSTTSTSdtststsTSe ne r gyc r os sSSSno r ms i gn alsssSSe ne r gys i gn alsssSv e c t ors i gn alnnnn????????????????????????????11,:),(),(...::...:],...,[:22112222121????For 2D vectors that are r rows by c columns, just concatenate the rows to make a single vector of size rxc. CVPR04 Stockman (27 June 04) 13 Equivalent function mathematics ),())()((),()(,)()(],[22fgdfggfdxxgxfgfddxxffSodxxgxfgfde f i n ec anbaonc on t i nu ou sbegandfl e tbababa????????????????????CVPR04 Stockman (27 June 04) 14 Cross correlate model (or mask) M with digital image I on neighborhood of m rows and n columns ],[],[),(1010rcMrycxIyxmrrncc??? ? ????????? The model is translated to location [x,y] ? m x n multiplications performed。 resulting scalar can be large ? doubly nested forloops to pute a single correlation ? quadruply nested forloops to create an image of cross correlations at “all” image locations [x,y] ? sometimes, mean is subtracted before correlation done CVPR04 Stockman (27 June 04) 15 Salient point selection ? perhaps has high energy: sum of the squares of all intensities is high ? perhaps variance of intensities is high ? perhaps variance in vertical, horizontal, and diagonal directions is high (Moravec interest operator example below) CVPR04 Stockman (27 June 04) 16 Interesting neighborhood de
教學(xué)課件相關(guān)推薦
鄂ICP備17016276號(hào)-1