Desktop version

Home arrow Engineering arrow Cellular Image Classification

Source

Introduction

The characterization of rotation invariant is significant for representing staining patterns of HEp-2 cell images, because the cells are not characterized by principal directions which are useful for aligning the HEp-2 cell images. To improve the classification performance, we propose two kinds of rotation invariant descriptors to characterize HEp-2 cells that are highly discriminative and descriptive with respect to their staining patterns.

We firstly propose a rotation invariant textural feature of PLTP-SRI. The Local Binary Pattern (LBP) operator had become well known as a simple and effective textural descriptor. In recent years, LBP-related features have been applied in stain-

© Springer International Publishing AG 2017 X. Xu et al., Cellular Image Classification, DOI 10.1007/978-3-319-47629-2_6

ing pattern classification of HEp-2 cells. For example, Nosaka et al. [1] utilize an extension of LBP descriptor, named Co-occurrence of Adjacent LBP (CoALBP), to extract textural features. CoALBP has a high descriptive ability and retains the advantages by considering the spatial relations among adjacent LBPs. The method won the first prize in HEp-2 cells classification contest. To further improve classification performance of CoALBP, Nosaka et al. [2] propose Rotation Invariant Co-occurrence among adjacent LBP (RIC-LBP) which is highly descriptive and invariant to image rotation. Our group [3] construct a dual spatial pyramid structure on a rotation invariant texture feature to take resolution variations into account and capture spatial layout information of the HEp-2 cell images. Theodorakopoulos et al. [4] propose a new descriptor based on Gradient-oriented Co-occurrences of LBPs (GoC-LBPs). Then it is fused with the distribution of SIFT features into a dissimilarity space. Nanni et al. [5] utilize a pyramid multi-scale representation coupled with a multiresolution LBP to represent the HEp-2 cell images.

However, the intensity gradients of our HEp-2 cells are weak, especially in the intermediate intensity cells, as shown in Fig. 6.1. LBP related features are sensitive to noise and smooth weak illumination gradients since they threshold at the gray value of the center pixel [6]. To solve the problem, we replace the binary patterns by three- value patterns which is more efficient than LBP for such a specific classification task. Furthermore, we propose a spatial pyramid structure based on patch-level rotation invariant Local Ternary Pattern (LTP) to capture spatial layout information.

Secondly, we apply a BoW framework for representing the HEp-2 cell images. The BoW framework is one of the most successful image classification framework [7, 8]. Our previous work verified its effectiveness in HEp-2 cell classification [9, 10]. It presents an image in terms of a set of visual words, selected from a trained beforehand codebook. By utilizing a BoW model, we can largely avoid the affection of position and orientation of images, and get a high discrimination rate. To enhance the discrimination power, we further add a spatial pyramid structure to retain spatial information.

As an optimization, the aforementioned two descriptors are integrated into a discriminative and descriptive image representation. Both features are respectively effective for capturing informative characteristics of the staining patterns in their

HEp-2 slide images with positive and intermediate fluorescent intensity

Fig. 6.1 HEp-2 slide images with positive and intermediate fluorescent intensity

  • 6.1 Introduction
  • 107

own ways. While our proposed PLTP-SRI feature extracts local feature, BoW builds a global image representation. It is reasonable to extract multiple features for compensation. The combined feature can take the advantages of the two kinds of features from different aspects. Furthermore, we demonstrate the validity of the proposed feature by experimental results consistently.

 
Source
Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >