Multi-dimensional empirical mode decomposition (EMD) method for image texture analysis
AssigneeIndustrial Technology Research Institute
InventorsSun-Hua Pao, Yio-Wha Shau, Ming-Chien Yu, Chien-Lun Tseng
Life of this patent
Estimating the term…
Abstract
A multi-dimensional empirical mode decomposition method is provided. The method can be applied in image texture analysis, such as medical image analysis. The method can adaptively decompose a three-dimensional image into a number of characteristic image layers and extract characteristic images showing more noticeable textures from the layers. The method uses the physical concept of field to perform the data mode decomposition to obtain envelope and tendency estimation of multi-dimensional data. The method can also be applied in time and frequency analysis of two-dimensional data or multi-channel data.
Claims30
A method for processing multi-dimensional data by empirical mode decomposition (EMD), comprising:
- reading a first set of multi-dimensional data;
- and
- sifting the first set of multi-dimensional data by iteration to obtain at least an intrinsic mode function via empirical mode decomposition, wherein the sifting step includes an iteration process for obtaining the intrinsic mode function comprising:
- mapping each value of the first set of the multi-dimensional data to a physical quantity of a physical field, described by a governing differential equation, to obtain a plurality of numerical values of the physical quantity, wherein the physical quantity has the same dimension as the first set of multi-dimensional data;
- determining an average envelope of a distribution of the numerical values of the physical quantity according to a varying relationship of the physical field with respect to the physical quantity;
- determining a next set of multi-dimensional data according to the first set of multi-dimensional data and the average envelope comprising:
- inversely mapping the average envelope in terms of the physical quantity to numeral values;
- and
- determining the next set of multi-dimensional data to be difference values between the first set of multi-dimensional data and the numerical values obtained by inversely mapping the average envelope;
- and
- performing a next iteration process starting with the mapping step with respect to the next set of multi-dimensional data.
A method for processing multi-dimensional image data by empirical mode decomposition, applied for image texture analysis, the method comprising:
- (a) reading a first set of multi-dimensional image data;
- (b) sifting the first set of multi-dimensional image data by iteration to obtain a plurality of intrinsic mode functions via empirical mode decomposition, wherein an iteration process of the sifting step for obtaining one of the intrinsic mode functions comprises:
- mapping each value of the first set of the multi-dimensional data to a physical quantity of a physical field to obtain a plurality of numerical values of the physical quantity, wherein the physical quantity has the same dimension as the first set of multi-dimensional image data and is a function of time;
- determining an average envelope of distribution of the numerical values of the physical quantity according to a varying relationship of the physical field with respect to the physical quantity;
- determining a next set of multi-dimensional image data according to the first set of multi-dimensional image data and the average envelope;
- and
- performing a next iteration process beginning with the mapping step according to the next set of multi-dimensional image data;
- and
- (c) transforming the intrinsic mode functions to be a plurality of multi-dimensional images and outputting the multi-dimensional images for the image texture analysis.
Dependencies are inferred from the claim text.