Towards computational models of kinship verification
read more
Citations
Neighborhood repulsed metric learning for kinship verification
Discriminative Deep Metric Learning for Face and Kinship Verification
Understanding Kin Relationships in a Photo
Discriminative Multimetric Learning for Kinship Verification
Prototype-Based Discriminative Feature Learning for Kinship Verification
References
LIBSVM: A library for support vector machines
Pictorial Structures for Object Recognition
Attribute and simile classifiers for face verification
Studying aesthetics in photographic images using a computational approach
Studying Aesthetics in Photographic Images Using a Computational Approach
Related Papers (5)
Frequently Asked Questions (15)
Q2. What future works have the authors mentioned in the paper "Towards computational models of kinship verification" ?
In order to develop more accurate kinship verification models, the authors plan on increasing the number of images in their ground truth database and conduct a larger scale human performance evaluation, both on kin relation verification ability and key inherited facial feature identification. Additional future work includes exploring genealogical models to characterize kinship, investigating where in face the cues that signal kinship falls by blacking out facial parts and showing the remaining face, and shedding light on which features of their kinship verification model may be universal vs. gender-dependent, and assessing the influence of age, race etc. in kinship verification so as to form a general classification model. Finally, the authors also plan to develop novel kinship verification and key inherited facial feature locating user interface to assist solving social problems of lost children searching and historical consanguinity identification.
Q3. Why is the feature vector restricted to 6 component?
The feature vector is restricted to 6 component due to relatively small the number of image pairs in ground truth (150) and in order to prevent overfitting.
Q4. How do the authors improve the kinship verification model?
In order to develop more accurate kinshipverification models, the authors plan on increasing the number of images in their ground truth database and conduct a larger scale human performance evaluation, both on kin relation verification ability and key inherited facial feature identification.
Q5. How do the authors classify the parent-child pairs?
To classify the parent-child pairs into two true and falsecategories, the authors first create training and test sets by pairing up images in their ground truth set.
Q6. How is the cost of placing a part at a given position calculated?
Using the distance function, the cost of placing a part at a given position is computed by taking the distance of that position from the average position of the part.
Q7. What is the commonly occurring color in the facial part?
For hair color, a sub-window of the top of the image was taken, and a mode filter is applied to this sub-window to obtain the most commonly occurring color in this region.
Q8. How is the facial image database collected?
The facial image database of parent-child pairs is collected through a controlled on-line search for images of public figures and celebrities and their parents or children.
Q9. How do the authors plan to solve social problems of lost children?
the authors also plan to develop novel kinship verification and key inherited facial feature locating user interface to assist solving social problems of lost children searching and historical consanguinity identification.
Q10. How do the authors solve the novel challenge?
The authors address the novel challenge by posing the problem as a binary classification task, and extracting discriminative facial features for this problem.
Q11. How do the authors find the discriminative inherited facial features?
In order to find the most discriminative inherited facial features, the authors have applied the forward selection methods to kinship verification.
Q12. What is the method used to train the classifier?
Using the extracted feature vectors, the authors calculate the differences between feature vectors of the corresponding parents and children, and apply K-Nearest-Neighbors and Support Vector Machine methods to train the classifier on these difference vectors, as well as a set of negative examples (i.e., image pairs of two unrelated people)
Q13. What are the results of the experiments?
From their experimental results, the authors have found that these facial features are able to effectively discriminate between related and unrelated parents and children.
Q14. What is the energy function for ourpictorial structures framework?
the authors can model the energy function for ourpictorial structures framework as follows:),(),(),( yxDistyxMatchyxE −=Where the Match function is the match score computed from normalized cross-correlation, and the Dist function is the distance from the position to the average part position.
Q15. How accurate is the classification accuracy of the parent-child dataset?
Finally the authors make a human performance evaluation on Parent-child dataset with classification accuracy of 67.19% on the whole database and varying accuracy across gender of parents and children.