A comparative study of image low level feature extraction algorithms

In this article, we present an extensive evaluation of visual descriptors for the content-based retrieval of remote-sensing (RS) images. The evaluation includes global hand-crafted, local hand-craf...

Visual descriptors for content-based retrieval of remote sensing images

In this paper we present an extensive evaluation of visual descriptors for the content-based retrieval of remote sensing (RS) images. The evaluation includes global hand-crafted, local hand-crafted, and Convolutional Neural Network (CNNs) features coupled with four different Content-Based Image Retrieval schemes. We conducted all the experiments on two publicly available datasets: the 21-class UC Merced Land Use/Land Cover (LandUse) dataset and 19-class High-resolution Satellite Scene dataset (SceneSat). The content of RS images might be quite heterogeneous, ranging from images containing fine grained textures, to coarse grained ones or to images containing objects. It is therefore not obvious in this domain, which descriptor should be employed to describe images having such a variability. Results demonstrate that CNN-based features perform better than both global and and local hand-crafted features whatever is the retrieval scheme adopted. Features extracted from SatResNet-50, a residual CNN suitable fine-tuned on the RS domain, shows much better performance than a residual CNN pre-trained on multimedia scene and object images. Features extracted from NetVLAD, a CNN that considers both CNN and local features, works better than others CNN solutions on those images that contain fine-grained textures and objects.

An efficient framework for image retrieval using color, texture and edge features

Worldwide, law enforcement agencies are encountering a substantial increase in the number of illicit drug pills being circulated in our society. Identifying the source and manufacturer of these illicit drugs will help deter drug-related crimes. We have developed an automatic system, called Pill-ID to match drug pill images based on several features (i.e., imprint, color, and shape) of the tablet. The color and shape information is encoded as a three-dimensional histogram and invariant moments, respectively. The imprint on the pill is encoded as feature vectors derived from SIFT and MLBP descriptors. Experimental results using a database of drug pill images (1029 illicit drug pill images and 14,002 legal drug pill images) show 73.04% (84.47%) rank-1 (rank-20) retrieval accuracy.

/pdf/pill-id-matching-and-retrieval-of-drug-pill-images-q-31jezd217z.pdf

Pill-ID: Matching and retrieval of drug pill images q

Color histograms lack spatial information and are sensitive to intensity variation, color distortion and cropping. As a result, images with similar histograms may have totally different semantics. The region-based approaches are introduced to overcome the above limitations, but due to the inaccurate segmentation, these systems may partition an object into several regions that may have confused users in selecting the proper regions. In this paper, we present a robust image retrieval based on color histogram of local feature regions (LFR). Firstly, the steady image feature points are extracted by using multi-scale Harris-Laplace detector. Then, the significant local feature regions are ascertained adaptively according to the feature scale theory. Finally, the color histogram of local feature regions is constructed, and the similarity between color images is computed by using the color histogram of LFRs. Experimental results show that the proposed color image retrieval is more accurate and efficient in retrieving the user-interested images. Especially, it is robust to some classic transformations (additive noise, affine transformation including translation, rotation and scale effects, partial visibility, etc.).

Jun-Feng Wu

Papers

Robust image retrieval based on color histogram of local feature regions