Adaptive Near Duplicate Image Retrieval Using SURF and CNN Features

Abstract

In this paper, we present an adaptive approach in order to match and retrieve near duplicate images at different scales. Matching only local Features does not necessarily identify visually similar images. Global features are fast at matching but less accurate. Many existing methods either use local features or CNN features for image or video retrieval task. In this paper, we combined the use of SURF local points and CNN features extracted around SURF points in order to match near duplicate image pairs. Image pairs are segmented into blocks and CNN features of the image block containing matched SURF features are extracted and matched. Regions around matched image blocks are grown adaptively and matching is carried out until CNN mismatch is observed. To verify our proposed approach, experiments are carried out on benchmarking California-ND and Holiday dataset. Compared to traditional approaches for image retrieval, our approach not only retrieves relevant images but also provides detail of localized matched patch. For California-ND dataset and Holiday dataset, we achieve remarkable mAP (mean average precision) score up to 0.86 and 0.74 respectively.

Near Duplicate Image Retrieval using Multilevel Local and Global Convolutional Neural Network Features

“…Smaller the block size, higher the number of CNN features and vice versa. However, in this experiment we followed the same block size mentioned in [9]. Performance improvement is found due to obtaining CNN for all selected local blocks as well as global (entire image).…”

“…2) Improvement in our previous approach [9]: We found that retrieving images based on considering only local image regions may not always give correct results. The proposed technique extracts local CNN features as well as global CNN features from images.…”

“…Later on, researchers found local features to be an effective way for various computer vision tasks and are more robust than global features. In multimedia retrieval, traditional local feature descriptors like Scale invariant feature transform-SIFT [9] and speeded up robust features-SURF [7] are popular. However, these feature descriptors may give false matching.…”

“…To summarize, the contribution of our work is as follows: 1) Pre-computed multi level local and global CNN block based matching: We focused on obtaining and storing CNN features of local blocks. This avoids extracting CNN features as and when there is a match, as mentioned in our earlier work [9]. This eliminates unnecessary overhead of repeatedly extracting CNN features from the same image.…”

CNN-Based Self Localization Using Visual Modelling of a Gyrocompass Line Mark and Omni-Vision Image for a Wheeled Soccer Robot Application

“…For Human Object Identification for Human-Robot Interaction, Fast R-CNN is used in [13], and hybrid Fuzzy-CNN in [14]. Besides, the implementation of CNN has been used for many applications, such as for matching near duplicate image [15], for situation prediction and sentiment analysis [16,17], for handwritten recognition [18], and for human brain segmentation in medical [19].…”