State-of-the-art image retrieval and image categorization systems achieve scalability by recent development of Bag-of-
Words model which is widely used as base image descriptor with good performance. Also, TF-IDF weighting scheme
achieves performance improvement of Bag-of-Words model, but their performance degrades significantly in large scale
system, due to low distinctiveness. We focus on boosting Bag-of-Words model performance with weighting scheme. We
present weighted similarity measure derived from word correlation for image categorization problem with small test
samples in each category, instead of weighting feature dimension directly. With deficient data sample to learn weight, our
method robustly calculate the weight elements. We show the proof of our similarity measure and its efficient performance
with UK Benchmark set.
Words model which is widely used as base image descriptor with good performance. Also, TF-IDF weighting scheme
achieves performance improvement of Bag-of-Words model, but their performance degrades significantly in large scale
system, due to low distinctiveness. We focus on boosting Bag-of-Words model performance with weighting scheme. We
present weighted similarity measure derived from word correlation for image categorization problem with small test
samples in each category, instead of weighting feature dimension directly. With deficient data sample to learn weight, our
method robustly calculate the weight elements. We show the proof of our similarity measure and its efficient performance
with UK Benchmark set.