================================================================================ SUPPLEMENTARY DATA for the research on FEATURE POINT DETECTION IN HIGH DYNAMIC RANGE IMAGERY ================================================================================ Date of last modification: 2015-12-15 Authors: Bronislav Přibyl, ipribyl@fit.vutbr.cz Alan Chalmers, alan.chalmers@warwick.ac.uk Pavel Zemčík, zemcik@fit.vutbr.cz Lucy Hooberman, l.hooberman@warwick.ac.uk Martin Čadík, cadik@fit.vutbr.cz Description: This data has been created when carrying out the research on "feature point detection in high dynamic range imagery" at the Brno University of Technology, Faculty of Information Technology, Czech Republic and the University of Warwick, Warwick Manufacturing Group, United Kingdom. If you use any part of this material in a publication, please cite our paper "PŘIBYL Bronislav, CHALMERS Alan, ZEMČÍK Pavel, HOOBERMAN Lucy and ČADÍK Martin. Evaluation of Feature Point Detection in High Dynamic Range Imagery. Journal of Visual Communication and Image Representation. Elsevier Science, 2016, vol. 2016, p. 20. ISSN 1047-3203.". BibTex entry: @ARTICLE{ author = {Bronislav P{\v{r}}ibyl and Alan Chalmers and Pavel Zem{\v{c}}{\'{i}}k and Lucy Hooberman and Martin {\v{C}}ad{\'{i}}k}, title = {Evaluation of Feature Point Detection in High Dynamic Range Imagery}, pages = {141 - 160}, journal = {Journal of Visual Communication and Image Representation}, volume = {38}, year = {2016}, ISSN = {1047-3203}, language = {english}, doi = {10.1016/j.jvcir.2016.02.007}, url = {http://dx.doi.org/10.1016/j.jvcir.2016.02.007} } The data contains images of two test scenes (2D and 3D) captured in three image sequences (distance-changing seq., lighting-chaning seq. and viewpoint-changing seq.) into 16 different image formats. Several single-exposure LDR images were captured at different exposure levels at each camera position and then combined into a single HDR image. The HDR image was subsequently tone mapped using 6 global and 5 local tone mappers. Six feature point (FP) detectors were tested: * Harris corner detector * Shi-Tomasi * Difference of Gaussians (a.k.a. DoG, the detector part of SIFT) * Fast Hessian (the detector part of SURF) * FAST * BRISK The 6 FP detectors were then carried out on the 16 described image formats which yielded 96 sets of feature points for each camera position. We then performed an automatic camera calibration from a dot pattern in each image sequence to obtain the camera parameters. Based on these parameters, we computed the geometric relations between individual views. In the case of the 2D scene, the geometric relation between the views is a planar homography automatically obtained upon camera calibration. In the case of the general 3D scene, the views are related by the more complicated concept of epipolar geometry. Without a 3D model of the scene, a 1:1 mapping between pixels is impossible. We solved this by manually annotating each view of the 3D scene sequences with a triangle net. The net contains cca 570 triangles which cover the same features in each view of the scene. Finally, the repeatability rate of the feature point detectors was evaluated for individual image formats and image sequences. The results can be obtain from our aforementioned paper. Directory structure: / +-2D/ ..................... - scene folder | +-distance/ ............. - sequence subfolder | | +-100/ ................ - image subsubfolder | | | +-pt/ .............. - subfolder with single-exposure images used to | | | | create the HDR image (images are geometricaly | | | | corrected using PT Lens) | | | | | | | +-R.txt ............. - camera rotation matrix with respect to the | | | | reference image - see Format 1 below | | | +-T.txt ............. - camera translation vector with respect to | | | | reference image - see Format 2 below | | | +-ROI.100.txt ....... - Region Of Interest (coordinates) for FP detection | | | | - see Format 3 below; 2D scene only | | | +-ROI.100.png ....... - Region Of Interest (binary mask) for FP detection; | | | | ; 2D scene only | | | +-ROIs.100.png ...... - Region Of Interest of shadows (binary mask); | | | | ; 2D scene only | | | +-ROIm.100.png ...... - Region Of Interest of midtones (binary mask); | | | | ; 2D scene only | | | +-ROIh.100.png ...... - Region Of Interest of highlights (binary mask); | | | | ; 2D scene only | | | | | | | +-100.hdr ........... - original calibrated HDR image | | | +-100.HDR.hdr ....... - HDR image, scaled into interval [0.0-256.0] | | | +-100.logHDR.hdr .... - HDR image after log. map., scaled into [0.0-256.0] | | | +-100.LDR.jpg ....... - best single exposure LDR image | | | +-100.LDR.histeq.jpg - LDR after histogram equalization | | | +-100.LDR.CLAHE.jpg . - LDR after CLAHE | | | +-100.gXXX.jpg ...... - original HDR after global tone mapping using XXX | | | +-100.lYYY.jpg ...... - original HDR after local tone mapping using YYY | | | | | | +-100.FFF_FP.DDD.txt - coords. of FPs (see Format 4 below) detected using | | | detector DDD in image format FFF | | | | | | | | +-103/ | | | +-... | | | | | +-109/ | | | +-... | | | | | +-122/ | | | +-... | | | | | +-147/ | | | +-... | | | | | +-197/ | | | +-... | | | | | +-297/ | | | +-... | | | | | +-H.100.103.txt ....... - homography matrix between views 100 and 103 - see | | | Format 5 below | | +- ... | | +-H.297.197.txt ....... - homography matrix between views 297 and 197 | | +-K.txt ............... - camera matrix for given image sequence - see | | Format 6 below | | | +-lighting/ | | +-001/... | | +-010/... | | +-011/... | | +-100/... | | +-101/... | | +-110/... | | +-111/... | | +-H.001.010.txt ....... - homography matrix between views 001 and 010 - see | | | Format 5 below | | +- ... | | +-H.111.110.txt ....... - homography matrix between views 111 and 110 | | +-K.txt ............... - camera matrix for given image sequence - see | | Format 6 below | | | +-viewpoint/ | +-00/... | +-01/... | +-02/... | +-03/... | +-04/... | +-05/... | +-06/... | +-07/... | +-08/... | +-09/... | +-10/... | +-11/... | +-12/... | +-13/... | +-14/... | +-15/... | +-16/... | +-17/... | +-18/... | +-19/... | +-20/... | +-H.00.01.txt ......... - homography matrix between views 00 and 01 - see | | Format 5 below | +- ... | +-H.20.19.txt ......... - homography matrix between views 20 and 19 | +-K.txt ............... - camera matrix for given image sequence | | +-3D/ +-distance/ | +-... | +-triangle_net_3D_distance.mat - triangle net covering images from the | | distance changing sequence (see Format 8 below) | +-... | +-lighting/ | +-... | | +-triangle_net_3D_lighting.mat - triangle net covering images from the | | lighting changing sequence (see Format 8 below) | +-... | +-viewpoint/ | +-... | | +-triangle_net_3D_viewpoint.mat - triangle net covering images from the | | viewpoint changing sequence (see Format 8 below) | +-... File formats: All numbers are floating point numbers (a.k.a. "floats") stored in a plain text format on 25 characters (aligned right, filled with spaces from left). The representation consists of: * an optional sign (-) * the mantisa - one digit before the decimal point - the decimal point (.) - several digits after the decimal point * the exponent sign (e) * the exponent - the sign (+ or -) - one to three digits Example floats: " -9.9993639314785610e-001", " 3.6202320949113728e+03". Entries of a matrix row are stored on one line of a text file. The entries are separated by a tab character( ). Example of a matrix row with 3 entries: " -9.9993639314785610e-001 -1.7425725739943524e-003 -1.1143298402206141e-002" == Format 1 == - rotation matrix - 3x3 matrix - each matrix row is on a separate line == Format 2 == - translation vector - 3x1 vector - each vector entry is on a separate line == Format 3 == - a region of interest (sequence of image coordinates defining a polygon; the first and last coordinate is NOT the same) - nx2 matrix - each coordinate pair (x y) is on a separate line == Format 4 == - coordinates of feature points - nx3 matrix - each feature point is on a separate line; first two entries are the (x y) coordinates, the last entry is the "strength" of the feature point (or the response of the feature point detector) == Format 5 == - homography matrix - the same as Format 1 == Format 6 == - camera matrix - the same as Format 1 - the meaning of individual matrix entries is explained at http://www.vision.caltech.edu/bouguetj/calib_doc/htmls/parameters.html (see 'camera matrix') == Format 7 == - fundamental matrix - the same as Format 1 == Format 8 == This is an exception - NOT a text file BUT a Matlab workspace file. There is one variable stored in it. The variable is a 3D matrix with triangle vertices. - TxIx6 matrix of doubles, where - T = number of triangles in the net, - I = number of images in the sequence (for example I=21 for viewpoint ch. seq.) - each 6-tuple along the third dimension contains the image pixel coordinates of vertices A,B,C of each individual triangle: [Ax Ay Bx By Cx Cy]