Rethinking Rotation-Invariant Recognition of Fine-grained Shapes from the Perspective of Contour Points

TL;DR

This work tackles rotation-invariant recognition of fine-grained shapes by shifting from pixel-based to contour-point representations. It introduces the Local Orientable Axis (LOA) and Local Orientable Axis Information (LOAI) encoded over Local Geometric Areas (LGAs), forming an anti-noise rotation-invariant convolution module that is cascaded into five layers. The method demonstrates strong rotation invariance and robustness to contour noise in both shape classification and retrieval tasks across Kimia99, Flavia, Kidney, and Liver datasets, outperforming several state-of-the-art pixel-based approaches. The approach enables end-to-end learning with end-to-end differentiability and shows promise for broader 2D signal applications, with publicly available code for replication.

Abstract

Rotation-invariant recognition of shapes is a common challenge in computer vision. Recent approaches have significantly improved the accuracy of rotation-invariant recognition by encoding the rotational invariance of shapes as hand-crafted image features and introducing deep neural networks. However, the methods based on pixels have too much redundant information, and the critical geometric information is prone to early leakage, resulting in weak rotation-invariant recognition of fine-grained shapes. In this paper, we reconsider the shape recognition problem from the perspective of contour points rather than pixels. We propose an anti-noise rotation-invariant convolution module based on contour geometric aware for fine-grained shape recognition. The module divides the shape contour into multiple local geometric regions(LGA), where we implement finer-grained rotation-invariant coding in terms of point topological relations. We provide a deep network composed of five such cascaded modules for classification and retrieval experiments. The results show that our method exhibits excellent performance in rotation-invariant recognition of fine-grained shapes. In addition, we demonstrate that our method is robust to contour noise and the rotation centers. The source code is available at https://github.com/zhenguonie/ANRICN_CGA.

Figures (17)

• Figure 1: The process and tricky scenario of shape recognition
• Figure 2: The overview of anti-noise rotation-invariant convolution module.
• Figure 3: The calculation process of the Local Orientable Axis(LOA).
• Figure 4: The encoding process of Local Orientable Axis Information(LOAI). (a) Local geometric areas are encoded into LOAI matrices. (b) Features of local geometric areas.
• Figure 5: Feature Enhancement Module. After the encoded LOAI matrices are combined and aligned, they pass through a two-layer multi-channel 1D convolutional kernel to achieve feature enhancement.