Probabilistic web image gathering
read more
Citations
A Framework for Web Science
OPTIMOL: Automatic Online Picture Collection via Incremental Model Learning
Dataset issues in object recognition
OPTIMOL: automatic Online Picture collecTion via Incremental MOdel Learning
Animals on the Web
References
Indexing by Latent Semantic Analysis
Term Weighting Approaches in Automatic Text Retrieval
Combining labeled and unlabeled data with co-training
Object class recognition by unsupervised scale-invariant learning
Object Recognition as Machine Translation: Learning a Lexicon for a Fixed Image Vocabulary
Related Papers (5)
Frequently Asked Questions (11)
Q2. What are the future works in "Probabilistic web image gathering" ?
As future work, the authors plan to apply the proposed method to automatic generation of real world image corpus for generic image classification/recognition. In their method, the authors have already made generative models based on the Gaussian mixture which can be applied to generic image recognition. So the authors need to develop new models which can represent “ object ” concepts as well as “ scene ” concepts by importing latest methods for generic object recognition.
Q3. What is the method to find regions related to a certain concept?
Their method to find regions related to a certain concept is an iterative algorithm similar to the expectation maximization (EM) algorithm applied to missing value problems.
Q4. What are the three kinds of features that the authors prepare for image features?
As image features, the authors prepare three kinds of features: color, texture and shape features, which include the average RGB value and its variance, the average response to the difference of 4 different combination of 2 Gaussian filters, region size, location, the first moment and the area divided by the square of the outer boundary length.
Q5. How do the authors make a word vector for each image?
To make a word vector for each HTML document, the authors eliminates HTML tags and extracts surrounding ten words (only nouns, adjectives, and verbs) before and after the link tag to the image file, link words and words in the ALT tag from HTML documents associated with downloaded images.
Q6. How do the authors use the Gaussian mixture model to represent models associated to keywords?
The authors use a generative model based on the Gaussian mixture model to represent models associated to keywords, and estimate models with the EM algorithm.
Q7. How many relevant images were needed to be supervised by hand?
In their paper, they claimed only 50 relevant images were needed to be supervised by hand in case of using colearning with both visual and textual features.
Q8. How many images are relevant to the keyword “X”?
In their previous work [16], the authors revealed that images whose file name, ALT tag or link word includes a certain keyword “X” are relevant to the keyword “X” with around 75% precision on average.
Q9. What is the contribution of this paper?
The contribution of this paper is as follows:(1) The authors divide raw images collected from the Web into two groups, A and B by analyzing associated HTML documents, and use images in group A which are more likely to be relevant as initial training images for the probabilistic learning method.
Q10. How do the authors compute the probability of “X” and “non-X” images?
To realize that, the authors compute the probability of “X” and “non-X” in terms of word vectors in the same way as image features, and integrate them.
Q11. What is the key improvement in gather?
The key improvement in their gather approach over earlier work [16, 17] is the construction of a probabilistic model for the relevant part of the images.