Majorppt
The document discusses content-based image retrieval (CBIR) using machine learning techniques. It describes using a support vector machine (SVM) with color histograms to classify images into categories during training. An optimal color histogram bin size is derived for each image to use as a feature vector, along with shape/texture features, for classification and retrieving similar images within categories. The approach is tested on 12 categories containing 720 training and 210 test images, extracting bag-of-features descriptors. SVM prediction achieves classification, then similar images are retrieved within categories based on optimal bin size. However, as data sets increase, unrelated images may be mapped to the same features, limiting generic CBIR until advances in image understanding and hardware. Focusing on
Majorppt
- 3. he amount of pictorial data has been growing enormously with the expansion of Web. From the large number of images, it is very important for users to retrieve required images via an efficient and effective mechanism. Most of the images are retrieved by text based image retrieval which is not very efficient. The images are indexed by tags in TBIR. For retrieving image the text entered by user matched with image tags. E.g. for searching a image one can miss the important text then the output come is not very desired and satisfactory for user.
- 4. Support Vector machine(SVM) with Color Histogram:-using train images where each image associate with one of the given categories. SVM with Optimal Histogram Bin SIZE(derived variable):- Optimal bin size is obtained for color histogram of each image. The optimal bin size is used for finding similar images within the category.
- 5. The optimized bin size for each image present in the training database is calculated. The bin size will be used as a feature The feature vector consists of two features:- The extracted shape/texture features(SIFT). Optimal bin size.
- 6. 720 images are present in the database for training. Algorithms applied on the images present in the database. Descriptor features like color histogram, SIFT, Bag of feature are used. There are total of 12 categories each containing 60 images. Categories chosen are aero plane, bus, flowers, car, bike etc.
- 7. For testing purpose we have taken 210 images. The bow descriptors are extracted for images of each category. The extracted descriptors are given to the SVM as an input. The SVM predict the class of the images. Further within the category the similar images are retrieved based on optimal bin size.
- 10. We find that the transformation from images to features (or other descriptors) is many-to- one and when the data set is relatively small, there are no collisions. But as the size of the set increases unrelated images are likely to be mapped into the same features.
- 11. We propose that generic CBIR will have to wait both for algorithmic advances in image understanding and advances in computer hardware. In the meantime we suggest that efforts should be focused on retrieval of images in specific applications where it is feasible to derive semantically meaningful features.