Spring 2020: Venu: Haag 312, Time: M/W 4-5:15pm ECE 5582 Computer Vision Lec 03: Image Formation - Geometry Zhu Li Dept of CSEE, UMKC Office: FH560E, Email: [email protected], Ph: x 2346. http://l.web.umkc.edu/lizhu Z. Li: ECE 5582 Computer Vision, 2020 p.1 slides created with WPS Office Linux and EqualX LaTex equation editor
39
Embed
ECE 5582 Computer Vision · Perspective Projection Examples Perspective projection is a simplification of real world image formation Lens characteristics are not considered Perspective
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Office: FH560E, Email: [email protected], Ph: x 2346.http://l.web.umkc.edu/lizhu
Z. Li: ECE 5582 Computer Vision, 2020 p.1
slides created with WPS Office Linux and EqualX LaTex equation editor
Outline
Recap of Lec 02 Projection Geometry of Image Formation Homography Summary
Z. Li: ECE 5582 Computer Vision, 2020 p.2
Outline
Recap of Lec 02 Projection Geometry of Image Formation Homography Summary
Z. Li: ECE 5582 Computer Vision, 2020 p.3
Demosaicing
Demosaicing in deep learning era: [5] Nai-Sheng Syu*, Yu-Sheng Chen*, Yung-Yu Chuang , “Learning Deep Convolutional
Networks for Demosaicing”.
Z. Li: ECE 5582 Computer Vision, 2020 p.4
handcrafted
deep learning based
DMCNN
Similar to our ICME work, residual learning blocks with BN and SE
Good performance:
Z. Li: ECE 5582 Computer Vision, 2020 p.5
paras' network [6] fordark image denoising and demosaic
HSV Color Model
Hue, Satuation and Value (brightness) color model
A cone with inituitive appeal of painters' tint, shade and tone model pure red: H=0, S=1, V=1. tints: adding white pigments, decreasing saturation shades: adding black, decrease brightness tones: dereasing S and V
Human can differentiate approx. 128 hues, and 130 levels of saturation
The number of values (brightness) is color dependent, approx 16 for blue, and 23 for yellow
Z. Li: ECE 5582 Computer Vision, 2020 p.6
MPEG-7 Scalable Color Descriptor
Scalable Color Descriptor:•Scalable Color Descriptor (SCD) is in the form of a color histogram in the HSV color space encoded using a Haar transform. H is quantized to 16 bin and S and V are quantized to 4 bins each, total 256 bins.
•The pixel count for each bin is quantized to 4 bits, so at max 256x4=1024 bits for representing. The distance between two images are therefore hamming distance, Scalability thru Haar trans.
Saturation
Hue
Value
Red (0o)
Yellow (60o)
Green (120o)
Cyan (180o)
Blue (240o) Magenta (300o)
Black
White
p.7Z. Li: ECE 5582 Computer Vision, 2020
Pooled Color Histogram
Color Histogram does not capture spatial info E.g, Frech and Czech flags will have the same descriptor.
Spatial pooling
Z. Li: ECE 5582 Computer Vision, 2020 p.8
Matlab Implementation
Very easy…
Z. Li: ECE 5582 Computer Vision, 2020 p.9
Adaptive Bin Color Histogram
SCD uses fixed color bins. This can be good: the feature is state-less,and can be easily
vectorized or hashed. But not as compact, nor as representative
How about let the color bins also adaptive to the images ?
Z. Li: ECE 5582 Computer Vision, 2020 p.10
3,4,..8 dominantColor approximation
Dominant Color Descriptor
Color Space Quantization K-Means clustering of colors into desired number of bins, n.
Specify an Affine matrix H, and then re-arrange pixels to simulate new image formation from different camera angles J = imwrap(I, A);
Z. Li: ECE 5582 Computer Vision, 2020 p.29
Why we care about Homography ? In object retrieval, we need to know exactly which interesting
points in one image, is related to the interesting points in another For re-ranking of short list For localizing the query object Depth estimation from homograph (non-GPS positioning)
Z. Li: ECE 5582 Computer Vision, 2020 p.30
Object re-identification: homography for geometry re-ranking
Navigation/non GPS positioning:
Homography Estimation – DLT Algorithm
To recover H, which has 8 DoF/variables, we just need 4 matching pixel pairs, ideally.
Say if we have 2 points on (u,v) and (x,y) plane matched
s��′�′1� = �
ℎ� ℎ� ℎ�ℎ� ℎ� ℎ�ℎ� ℎ� ℎ�
� ���1�
For each pair, rearranging by dividing first row with the 3rd, and the 2nd row with the 3rd, we have the following,
Do this for 4 points pair: (x1, y1; x’1, y’1),…, (x4, y4; x’4, y’4), we will have 8 equations to solve 8 parameters