Scribd
Upload
36 views3 pages

DIP Final

The document discusses various computer vision and deep learning concepts including Gabor filters, SIFT, convolutional and pooling layers in CNNs, AlexNet, VGG, ResNet, R-CNN, Fast R-CNN, Faster R-CNN, one-stage and two-stage detectors, semantic, instance and panoptic segmentation, CNNs and FCNs, object detection and tracking, facial landmarks, and applying deep learning to fingerprint verification.

Uploaded by

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

DIP Final

The document discusses various computer vision and deep learning concepts including Gabor filters, SIFT, convolutional and pooling layers in CNNs, AlexNet, VGG, ResNet, R-CNN, Fast R-CNN, Faster R-CNN, one-stage and two-stage detectors, semantic, instance and panoptic segmentation, CNNs and FCNs, object detection and tracking, facial landmarks, and applying deep learning to fingerprint verification.

Uploaded by

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

🏙

DIP Final
2021 exam

1. Explain why and how Gabor filter can be used to enhance fingerprint images
(<300 words)
A Gabor filter is just to enhance fingerprints by seeing which direction the
fingerprint is orientated in which direction to get the characteristics of the
fingerprint line to be clear; this is because Gabor filters tend to give out the
direction to be clear.

2. Explain the concept of SIFT keypoint detector and SIFT descriptor (<300 words)
SIFT keypoint detector and SIFT descriptor is a part of Scale-Invariant Feature
Transform, which SIFT keypoint detector is the first part which aims to get the
key point of the image by using convolute image at different scales. The
descriptor part is use to detect the image is same as each other by Compute the
gaussian function or Difference-of-Gaussian (DoG)

3. What are the purposes of convolutional layer and pooling layer in convolutional
Neural networks? (<200 words)
The convolutional layer is the part to extract the feature of the image by applying
a set of filters to the image and the pooling layer aims to reduce the amount of
data feeding to the subsequence layers by using max-pooling or average-
pooling filters.

DIP Final 1
4. Explain the differences between AlexNet, VGG, and ResNet (<300 words)
AlexNet is the first to introduce CNN and output like a global filter.VGG is the
improvement of the AlexNet by adding more layers and using a stack of small
filters (3×3) instead of a single large filter to improve the speed. ResNet
introduced a concept of skipping the layer because the ReesNet research found
out that deeper the network does not improve the performance and fix the
degradation problem. In making the layer able to skip, so it has the choice to
skip not just pass down the weight.

5. Explain the differences between R-CNN, Fast R-CNN, and Faster R-CNN (<300
words)
R-CNN is the first to introduce and it is the slowest because it just uses a
different model in its architecture of extract regions which use Selective search
to output image, which warps to be passed to CNN features that use AlexNet
and pass to Binary-SVms to Classify regions. Next is Fast R-CNN, it is faster
than R-CNN because it uses CNN and NN for Feature computation and Object
classification so it can share computation. Fast R-CNN also feeds the whole
image all at once and uses an ROI pooling layer that can handle variable-size
inputs by using Max pool. Last is Faster R-CNN, which is faster than Fast R-
CNN because it replaced Selective search to be RPN so now it can be
computed end to end. The way that Faster R-CNN is trained is different from the
others too.

6. Explain the concept of one-stage and two-stage detectors (<200 words)


One-stage is computer localization and classification at the same time like YOLO
and SSD, but for two-stage detectors localization and classification is done
separately like R-CNN, Fast R-CNN, and Faster R-CNN .

7. Explain the differences between semantic segmentation, instance segmentation,


and panoptic segmentation (<200 words)
Panoptic segmentation is the combination of semantic segmentation, instance
segmentation

8. What are convolutional neural networks (CNNs) and fully-convolutional networks


(FCNs)? (<300 words)
CNNs is a class of feedforward neural network, which has two parts: to extract
the feature of the image and the other is to classify what the thing that got feed
in the feature learning is.
FCNs are the same as CNNs, but the layers that flatten to be convolution layers
instead of fully-connected layers. FCNs also use the concept of Deconvolution to
make the image bigger by adding padding.

DIP Final 2
9. What are the differences between object detection and object tracking? (<200
words)
Object detection is to classify what the object is. And object tracking is to identify
the movement of the object.

10. What are facial landmarks and how are they applied in a face recognition
system?
(<200 words)
Facial landmarks are the main part of the eyes, mouth, nose, and eyes brow. It is
use in face recognition systems by recognizing the disteness on the facial
landmarks.

11. Is it possible to apply a deep neural network for a fingerprint verification system?
If
so, explain the model architecture and how to train the model. If not, why? (<500
words)
Yes it is possible to apply a deep neural network for fing for a fingerprint
verification system. It can use Local binary patterns( LBP ) for detecting the
texture of the finger print first then feed it to neural networks to detect
fingerprints.

Fingerprint verification

📽 Image features and descriptors


🏝 ML and Artificial neural networks
♟ Convolution Neural Network
Deep learning part1

🏎 Deep learning part 2


🤢 Image segmentation using deep learning
Cheat sheet

DIP Final 3

You might also like