Journal Article
Computer Vision

Discovering Discriminative Graphlets for Aerial Image Categories Recognition

Luming Zhang(Zhejiang University), Yahong Han(Tianjin University), Yi Yang(Queensland University of Technology), Mingli Song(Zhejiang University), Shuicheng Yan(National University of Singapore), Qi Tian(The University of Texas at San Antonio)
August 14, 2013IEEE Transactions on Image Processing235 citations
DOI

235

Citations

4

Influential Citations

IEEE Transactions on Image Processing

Venue

2013

Year

Abstract

Recognizing aerial image categories is useful for scene annotation and surveillance. Local features have been demonstrated to be robust to image transformations, including occlusions and clutters. However, the geometric property of an aerial image (i.e., the topology and relative displacement of local features), which is key to discriminating aerial image categories, cannot be effectively represented by state-of-the-art generic visual descriptors. To solve this problem, we propose a recognition model that mines graphlets from aerial images, where graphlets are small connected subgraphs reflecting both the geometric property and color/texture distribution of an aerial image. More specifically, each aerial image is decomposed into a set of basic components (e.g., road and playground) and a region adjacency graph (RAG) is accordingly constructed to model their spatial interactions. Aerial image categories recognition can subsequently be casted as RAG-to-RAG matching. Based on graph theory, RAG-to-RAG matching is conducted by comparing all their respective graphlets. Because the number of graphlets is huge, we derive a manifold embedding algorithm to measure different-sized graphlets, after which we select graphlets that have highly discriminative and low redundancy topologies. Through quantizing the selected graphlets from each aerial image into a feature vector, we use support vector machine to discriminate aerial image categories. Experimental results indicate that our method outperforms several state-of-the-art object/scene recognition models, and the visualized graphlets indicate that the discriminative patterns are discovered by our proposed approach.

Analysis

Why This Paper Matters

This paper addresses a critical gap in aerial image recognition: the inability of generic visual descriptors to capture the geometric layout and spatial relationships among components like roads and playgrounds. By introducing graphlets—small connected subgraphs that encode both topology and appearance—the authors enable a more discriminative representation for aerial scene categories. This is particularly significant for applications such as urban planning, environmental monitoring, and surveillance, where understanding spatial structure is key.

The work bridges graph theory and computer vision, offering a principled way to compare region adjacency graphs through graphlet matching. The manifold embedding step is crucial for handling graphlets of varying sizes, a common challenge in graph-based methods. This approach moves beyond bag-of-features models by explicitly modeling spatial interactions, which are often lost in local feature aggregation.

Technical Contributions

  • Graphlet-based representation: Defines graphlets as small connected subgraphs of region adjacency graphs, capturing both geometric and appearance information.
  • RAG construction: Decomposes aerial images into basic components (e.g., roads, playgrounds) and builds a region adjacency graph to model spatial interactions.
  • Manifold embedding for graphlet comparison: Derives an algorithm to measure similarity between graphlets of different sizes, enabling effective matching.
  • Discriminative graphlet selection: Selects graphlets with high discriminative power and low redundancy, reducing computational overhead while maintaining accuracy.
  • SVM classification: Quantizes selected graphlets into feature vectors and uses support vector machines for final category recognition.

Results

The method outperforms several state-of-the-art object/scene recognition models on aerial image categories. Visualized graphlets demonstrate that the discovered patterns are indeed discriminative, confirming the effectiveness of the approach. The paper reports experimental results but does not provide specific numerical metrics (e.g., accuracy percentages) in the abstract.

Significance

This work advances aerial image analysis by providing a geometric-aware representation that captures spatial layout, a dimension often overlooked by generic descriptors. It opens avenues for integrating graph-based reasoning into scene recognition, with potential extensions to other domains like medical imaging or remote sensing. The manifold embedding technique for comparing variable-size graphlets is a methodological contribution that could benefit graph-based learning more broadly.

aerial-image-recognitiongraphletsregion-adjacency-graphmanifold-embeddingsupport-vector-machinescene-annotationsurveillance
Quantization

Related Papers