Everyday concept detection in visual lifelogs: validation, relationships and trends
read more
Citations
The Evolution of First Person Vision Methods: A Survey
Passively recognising human activities through lifelogging
Toward Storytelling From Visual Lifelogging: An Overview
Remembering through lifelogging: A survey of human memory augmentation
Unconscious emotions: quantifying and logging something we are not aware of
References
The measurement of observer agreement for categorical data
LIBSVM: A library for support vector machines
The Nature of Statistical Learning Theory
A new method for gray-level picture thresholding using the entropy of the histogram
Related Papers (5)
Frequently Asked Questions (12)
Q2. What have the authors stated for future works in "Everyday concept detection in visual lifelogs: validation, relationships and trends" ?
The authors plan to undertake evaluations on lifelog collections to assess the utility of such retrieval methods. Finally, the authors believe that the exploration of active learning approaches which would combine user-contributed tagging ( or folksonomies ) with concept detection training, could be undertaken.
Q3. What are the main requirements for semantic concept detection on visual lifelogs?
The major requirements for semantic concept detection on visual lifelogs are as follows: a) the identification of everyday concepts; b) the identification of positive and negative examples; and c) reliable and accurate detection.
Q4. What is the main reason why SenseCam images are of low quality?
SenseCam images tend to be of low quality owing to: their lower visual resolution; their use of a fisheye lens which distorts the image somewhat but increases the field of view; and the absence of a lens aperture resulting in many images being much darker or brighter than desired for optimal visual analysis.
Q5. How do the authors construct a codebook model from a Gabor filter?
In order to obtain an image region descriptor with Gabor filters the authors follow these three steps: 1) parameterise the Gabor filters, 2) incorporate colour invariance, and 3) construct a histogram.
Q6. What are the sensors that are used to capture the environment of the wearer?
These are: a three-axis accelerometer - to detect movement of the wearer; a passive infrared sensor - to detect bodies in front of the wearer; light sensor - to detect changesin light level such as when moving from indoors to outdoors; and an ambient temperature sensor.
Q7. How can the performance of a concept detection be boosted?
If a semantic concept or query is known to be highly temporally consistent, then by consulting the prediction on the previous shot, the overall performance of concept detection can be boosted.
Q8. What is the common method used in video retrieval?
One such method is concept detection, an often employed approach in video retrieval [27,32,35], which aims to describe visual content with confidence values indicating the presence or absence of object and scene categories.
Q9. What are the advantages of concept-based retrieval?
As in video retrieval, these concepts offer the ability to bridge semantic understanding to enable search and location of images relevant to an information need.
Q10. How can the authors model the full range of image statistics in natural textures?
It was shown in [15] that the complete range of image statistics in natural textures can be well modeled with an integrated Weibull distribution, which in turn can be characterised by just 2 parameters.
Q11. What are the main issues that would be attributed to the poor performance of an everyday detector?
In summary, the authors would attribute poor performance of an everyday detector to one or more of the the following issues: a sub-optimal number of positive examples provided for training; a sub-optimal distribution of examples across the user’s collection; and/or a sub-optimal diversity in the visual distinctiveness of the provided positive examples (i.e. many highly visually similar examples).
Q12. How many unique images were judged by the 9 annotators?
This resulted in almost 1400 positive and negative unique images per concept to be judged by the 9 annotators (50 to be judged by all 9 plus 9×150 individual judgments).