automatic image registration

Upload: vihang-eth

Post on 04-Apr-2018

222 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/29/2019 Automatic Image Registration

    1/10

    Prepared By:

    Trivedi Vihang Harishchandra

    ROLL No: 28ME(MSA)

    Electrical Engg. Department

    Guided By:Prof. Satish K. Shah

    Head Of the Dept.Electrical Engg. Dept.FTE,Vadodara

  • 7/29/2019 Automatic Image Registration

    2/10

    Work Done

  • 7/29/2019 Automatic Image Registration

    3/10

    Input image(Unregistered

    )

    Divide Imageinto non

    overlappedblocks

    Cross-correlatedimages of those

    blocks withreference image

    Referenceimage

    Energies ofall images

    Sorting of allvalues ofenergies

    Select lowerenergyblocks

    Findmaximumcorrelated

    pixelcoordinate

    Final matchpoints.

    Works too well with lowerdistortion.

    But when distortion ismuch higher or image has

    too much edges, lines andpoints , Match point aretoo bad.

    NEED TO UNDERSTANDTHE FEATURES, SO THISREQUIRES CORRECT F/B

    at CRITICAL STAGES.

  • 7/29/2019 Automatic Image Registration

    4/10

    Input image(Unregistered

    )

    Divide Imageinto non

    overlappedblocks

    Cross-correlatedimages of those

    blocks withreference image

    Referenceimage

    Energies ofall images

    Sorting ofall values of

    energies

    Select lower energyblocks(N)

    5

  • 7/29/2019 Automatic Image Registration

    5/10

    DECISION CORE

    From sorted value. Take the maximum single crosscorrelation value, assign its 20% less as cutoff

    Watch on block numbers and cutoff.Give Template size as 32X32 for first iteration

    16X16 for second iteration for same number of blocks.

    Take i/p from those control points cross correlationvalue. If maximum value doesnt match cut-off thenselect next blocks. Blocks number should be less then

    80% of total blocks .

    If it matches the criteria make pair of 4 coordinates.Which are called control point pairs.

  • 7/29/2019 Automatic Image Registration

    6/10

    50 100 150 200 250 300 350 400 450 500

    50

    100

    150

    200

    250

    50 100 150 200 250 300 350 400 450 500

    50

    100

    150

    200

    250

  • 7/29/2019 Automatic Image Registration

    7/10

    0.83

    32

    16

    0.84

    32

    16

    0

    10

    20

    30

    40

    Cut-off TEMPLATESIZE

    BLOCK SIZE

    TRUEMATCH

    FALSEMATCH

    10

    23

    21

    13

    0

    10

    2030

    40

    50

    60

    70

    80

    BLOCK NUMBER

    NOT SHOULD BEUSED

    FALSE Match

    TRUE Match

    NOT MATCHED

  • 7/29/2019 Automatic Image Registration

    8/10

    50 100 150 200 250 300 350 400 450 500

    50

    100

    150

    200

    250

    50 100 150 200 250 300 350 400 450 500

    50

    100

    150

    200

    250

  • 7/29/2019 Automatic Image Registration

    9/10

    0.806

    32

    16

    0.88

    16 16

    0

    10

    20

    30

    40

    Cut-off TEMPLATE SIZE BLOCK SIZE

    TRUE MATCH

    FALSEMATCH

    29

    12

    12

    11

    0

    10

    2030

    40

    50

    60

    70

    BLOCK NUMBER

    NOT SHOULD BEUSED

    FALSE Match

    TRUE Match

    NOT MATCHED

  • 7/29/2019 Automatic Image Registration

    10/10

    Work Remaining:

    To develop transformation from achieved control points. To study Hardware (6748 and OMAP138) and implement this algorithm onsuitable platform.