Scribd
Upload
53 views3 pages

Encoding Techniques HOG Encoder

The document describes the HOG encoder technique in 3 steps: 1. Compute the gradient orientation and magnitude of pixels in an image and split it into cells. 2. Form histograms of gradient orientations within each cell and group cells into blocks to form the feature descriptor. 3. Normalize the blocks of histograms to account for illumination changes and represent the encoded image.

Uploaded by

sviji
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
53 views3 pages

Encoding Techniques HOG Encoder

The document describes the HOG encoder technique in 3 steps: 1. Compute the gradient orientation and magnitude of pixels in an image and split it into cells. 2. Form histograms of gradient orientations within each cell and group cells into blocks to form the feature descriptor. 3. Normalize the blocks of histograms to account for illumination changes and represent the encoded image.

Uploaded by

sviji
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Encoding Techniques HOG Encoder

Step 1: Compute the gradient for the pixels in the image.


Dx = [−1, 0, 1] and Dy = [−1, 0, 1] T
Gy x, y = I x, y ∗ Dy
Gx x, y = I x, y ∗ Dx

Gmag x, y = Gx2 x, y + Gy2 x, y

Gy x, y π
θ x, y = arctan +
Gx x, y 2
Step 2: Formation of cells and computation of histogram of oriented gradients.
Split each cell into angular bins according to the gradient orientation.
Each cell's pixel contributes weighted gradient to its corresponding angular
bin.
Step 3: Formation of cells into blocks for normalization.
Groups of adjacent cells - Blocks. The set of these block histograms represents the
1
feature descriptor.
Recognition Rate Comparison for Different Approach
using Extended Yale-B Database
Recognition Maximum
Author
Algorithm Recognition Rate
LBP 72%
Zhou et al. Gabor+ LBP 74.7 %
(2013) Gabor +LPQ 88.8 %
Gabor +LBP+ LPQ 90.7 % No. of Persons classified
Classification Accuracy =
LBP + PCA 88% Total No. of Persons
Proposed IPI ECG + PCA 90%
method ECG + LBP + PCA 98%

Author Recognition Algorithm Recognition Rate


Feng et al. (2000) Wavelet face + LDA 84%
Harihara & Gopala
HOG + SVM 88.6%
The proposed IPI is also found to
Krishna (2016) provide a correct estimation of age
Zhou et al. (2013) Gabor +LBP+ LPQ 90.7% under various illumination and
Mandal et al. (2009) Curvelet face + PCA + LDA 92%
low-resolution conditions.

Zhou et al. (2014) PCA and logistic regression 93.33%

Proposed IPI method ECG + LBP + PCA 98%

2
Conclusion
• Face Recognition and Age estimation
Face Recognition for uncontrolled lighting conditions is proposed.
Fusion-based Recognition technique also Estimate the age of the recognized
person and the accuracy is improved for low resolution images.
• Facial Expression Recognition
The automatic facial expression classification provides 90.00% for JAFFE
database which is better accuracy with lesser computational time.
• Heterogeneous Face Recognition
CFDA (Common Feature Discriminant Analysis) gives the highest
Recognition performance among the Encoding proposed methods .
• Wavelet for Face Recognition
A new VLSI architecture for 2-D SMDWT is Implemented which is superior
to the LDWT.
This Research work focused on illumination, pose and scale invariant algorithm
compared with existing method to evaluate the accuracy.
3

You might also like