CBIR based on linear SPM using SIFT sparse codes
K Bhargavi, Chava Santhi Rani · 2016
In today's present image processing world, Object Recognition is most important feature for images. In such signal and image processing sections Scale Invariant Feature Transform (SIFT) plays an important role as robust local invariant feature descriptor for object recognition. To further improve the recognition and subsequently the retrieval, Sparse coding using SIFT features are performed using a novel approach as Linear Spatial Pyramid Matching which improves the object recognition such as images or data. The content of the image is extracted using SIFT. Sparse feature representations for the SIFT features are then computed using the learned dictionary. These representations are further pooled via spatial pyramid matching kernels which have a fixed dimension feature vectors that represents the whole image. The SPM is used to maximum pool the local features from the images and improves the spatial information. The proposed method effectively represents the features of the images and these features are used in image retrieval. Based on the query image, the Euclidean distance is calculated with database images and similar images are resulted. And the Precision and Recall measures shows improved performance.