bag of words for classification,[url=http://www.hermesbirkin-bags.us@#$%&/]hermes belt[/url]
extract features from small image patches and then apply BOW onAfter the feature extraction I tried PCA,[url=http://www.louisvuittonoutlet-online.us/]cheap louis @#$%&[/url], feature selection,[url=http://www.louis-vuittonhandbags.co.uk/]louis @#$%& handbags[/url], changing no of clusters for KMeans etc to improve the accuracy. But in my case BOW learned on pixel values (1) outperforms (90%) than the BOW learned on features(2) (70%). My features are good and when I use to classify the images using some other framework I was able to get more than 95% accuracy.
My question is why BOW learned on pixels performs better than BOW learned on features,[url=http://www.ghd-hairstraighteners.co.uk/]ghd hair[/url]?
In "Decomposing a Scene into Geometric and Semantically Consistent Regions." (Gould et. al.) Shape,[url=http://www.prom-weddingdresses@#$%&/]prom dresses[/url], color,[url=http://www.monclerjackets-outlet@#$%&/]moncler outlet[/url], edge,[url=http://www.hermes-sac.fr/]hermes[/url], etc. features are derived by doing things like training boosted logistic regression classifiers,[url=http://www.louboutin-outlet.us@#$%&/]louboutins[/url], Potts models,[url=http://www.asicso@#$%&/]asics shoes[/url], and Gaussian Mixture models.
You probably don't need such intensive techniques to extract features that beat pixel vectors,[url=http://www.canadagoose.cc/]@#$%& coats[/url], but you should definitely browse around the literature to see what is effective.
SIFT features,[url=http://www.louboutinsit@#$%&/]louboutin[/url], color histograms,[url=http://www.moncleroutlet-jackets@#$%&/]moncler[/url], and filters to extract texture responses seem to work pretty well and also have a reasonable amount of software library support.
Thank you for the information. I tried 2: (1) extracting features (texture,[url=http://www.raybansunglasses.us@#$%&/]ray-ban sunglasses[/url], color,[url=http://www.michaelkorshandbags-uk.co.uk/]michael kors bags[/url], edge,[url=http://www.ferragamo-shoes-outlet@#$%&/]ferragamo belt[/url], etc) (2) only color. After extracting features I tried PCA/feature selection on the extracted features/intensity values. I found on some papers saying that for intensity values having a few hundred clusters are enough but for features we should have thousands of clusters. So currently I am trying with many clusters and hope it will give me good results. I will update the results soon.

[url=http://www.bjtsdhh@#$%&/bbs/read.php?tid-1220107.html]longchamp handbags Bag with Rs 4 lakh snatched[/url]

[url=http://www.sdrl.net/lt/boke.asp?lilhyi6bm.showtopic.48763.html]marc jacobs handbags Airport Safety[/url]

[url=http://www.md-zw@#$%&/home.php?mod=space&uid=13729&do=blog&quickforward=1&id=65883]juicy couture bags in the news[/url]

[url=http://www.morigele@#$%&/bbs/read.php?tid=10085568]@#$%& mercurial Biker boss jailed for two years[/url]

[url=http://www.think-group.cn/VK/blog/article.php?type=blog&cid=4&itemid=1712380]lululemon 'Dead' man who kicked out of a body bag has officially died[/url]