CN110414531A - SAR image Local Feature Extraction based on gradient ratio - Google Patents

Abstract

SAR image Local Feature Extraction based on gradient ratio, comprising the following steps: 1, calculate the gradient Ratio Features value GRP of original SAR image all pixels first；2, the binaryzation coded strings for generating its respective range neighborhood to each pixel using discriminant function, as the partial gradient ratio binary pattern for describing its characteristic；It is then combined with the binary-coding mode after rotating with identical minimal mode, its value is calculated, obtains the partial gradient ratio histogram of original SAR image；4, the characteristic similarity for comparing different images using symmetrical KL criterion SKLD, as the measurement to Feature Descriptor accuracy；Feature extraction algorithm proposed by the present invention can not only overcome the influence of SAR image multiplicative noise, and have good robustness to the rotationally-varying of target.

Description

SAR image Local Feature Extraction based on gradient ratio

Technical field

The present invention relates to electronic information technical fields, and in particular to a kind of SAR image local based on gradient ratio is special Levy extracting method.

Background technique

Similarity measurement between image can be measured by constructing SAR image Feature Descriptor, domestic and foreign scholars The related algorithm of Feature Descriptor is had conducted extensive research.In optical imagery field, local binary patterns (local Binary pattern, LBP) it is a kind of texture algorithm for describing central pixel point and surrounding pixel point gray scale size relation, it should Method calculates simple and has many advantages, such as part scale, rotation and bright dark invariance, be widely used in facial image analyze, The fields such as object detecting and tracking.In SAR image field, partial gradient ratio histogram (local gradient ratio Pattern histogram, LGRPH) it is that a kind of ratio estimates operator, overcome the coherent speckle noise and part ladder of SAR image Degree variation, is efficiently used for SAR image target identification.Since LBP operator is calculated based on Pixel gray difference, in SAR image In influence vulnerable to multiplicative noise, poor accuracy.And LGRPH feature lacks the description to orientation angle, to mesh in SAR image Mark rotationally-varying not robust.

Summary of the invention

The object of the present invention is to provide the SAR image Local Feature Extractions based on gradient ratio.

The problem to be solved in the present invention be LBP operator in SAR image vulnerable to the influence of multiplicative noise, poor accuracy, and LGRPH feature lacks the description to orientation angle, in SAR image the problem of not robust rotationally-varying to target.

To achieve the purpose of the present invention, the technical solution adopted by the present invention is that:

SAR image Local Feature Extraction based on gradient ratio, comprising the following steps:

1, the gradient Ratio Features value GRP of original SAR image all pixels is calculated first；

2, the binaryzation coded strings that generate its respective range neighborhood to each pixel using discriminant function, as describing its characteristic Partial gradient ratio binary pattern；

It is then combined with the binary-coding mode after rotating with identical minimal mode, its value is calculated, obtains original SAR image Partial gradient ratio histogram；

4, the characteristic similarity for comparing different images using symmetrical KL criterion SKLD, as the weighing apparatus to Feature Descriptor accuracy Amount.

The invention has the advantages that passing through the gray value difference of neighborhood territory pixel and center pixel in step 1 and neighborhood territory pixel ash Angle value makees ratio, obtains the noise immunity to SAR image, according to having identical minimum binary mode after cyclic shift in step 3, into Row coding mode merges, it is made to have rotational invariance, and step 4 calculates the similitude of the Feature Descriptor of different images, thus Obtain the ability for quickly carrying out SAR image recognition.

Detailed description of the invention

Fig. 1 is LBP operator calculating process；

Fig. 2 is LGRP feature calculation process；

Fig. 3 is the mapping relations of merging patterns；

Fig. 4 is similarity calculation process.

Specific embodiment

The present invention is further illustrated with reference to the accompanying drawings and embodiments.

The present invention is based on the SAR image Local Feature Extractions of gradient ratio, comprising:

1, the gradient Ratio Features value GRP of original SAR image all pixels is calculated first；Some picture of original SAR image is taken first The neighborhood of element calculates neighborhood territory pixel gradient value；

A) neighborhood territory pixel gradient value and itself gray value are made into ratio, obtains the gradient Ratio Features value of the neighborhood territory pixel；

B) average value for seeking the gradient Ratio Features value of all neighborhood territory pixels, the gradient Ratio Features value as the pixel；

C) thinking that binary-coding mode is then calculated according to LBP operator, calculates two of the pixel based on gradient Ratio Features value It is worth coding mode, as the description to the pixel；

D) coding mode for having identical minimum binary mode after cyclic shift is merged, obtained new model conduct pair The description of this kind of pixels；

E) characteristic value of the value of new model after merging as the pixel is calculated；

F) all pixels are successively traversed, the characteristic value of all pixels is obtained；

The characteristic value for counting all pixels forms grey level histogram, is exactly the invariable rotary partial gradient ratio of original SAR image Histogram (local gradient ratio pattern histogram, LGRPH).

2, the binaryzation coded strings for generating its respective range neighborhood to each pixel using discriminant function, it is special as it is described The partial gradient ratio binary pattern of property；

3, it is then combined with the binary-coding mode after rotating with identical minimal mode, its value is calculated, obtains original SAR image Partial gradient ratio histogram；

4, the characteristic similarity for comparing different images using symmetrical KL criterion SKLD, as the weighing apparatus to Feature Descriptor accuracy Amount；The feature vector for calculating two width figures, is denoted as respectivelyWith；

According to the definition of SKLD criterion, calculate the similarity of two width figures；

Using Gauss subordinating degree function, willValue range be mapped to SimilarityBetween value range.

Claims (1)

1. the SAR image Local Feature Extraction based on gradient ratio, it is characterized in that: the following steps are included:

The gradient Ratio Features value GRP of original SAR image all pixels is calculated first；

The binaryzation coded strings for generating its respective range neighborhood to each pixel using discriminant function, as the office for describing its characteristic Portion's gradient ratio binary pattern；

It is then combined with the binary-coding mode after rotating with identical minimal mode, its value is calculated, obtains original SAR image Partial gradient ratio histogram；

The characteristic similarity for comparing different images using symmetrical KL criterion SKLD, as the measurement to Feature Descriptor accuracy.