What are the algorithms for image feature extraction?

Commonly used image features include color features, texture features, shape features and spatial relationship features.

Color feature

(1) Feature: Color feature is a global feature that describes the surface properties of the scene corresponding to the image or image area. Generally speaking, color features are based on pixel features, and all pixels belonging to an image or an image area have their own contributions. Because color is insensitive to the change of the direction and size of an image or an image area, color features can't capture the local features of objects in the image well. In addition, when only using color features to query, if the database is very large, many unnecessary images will often be retrieved. Color histogram is the most commonly used method to express color characteristics. Its advantage is that it is not affected by image rotation and translation changes, and it is not affected by image scale changes by normalization. Its basic disadvantage is that it does not express the information of color space distribution.

(2) Common feature extraction and matching methods.

(1) color histogram

Its advantage is that it can simply describe the global distribution of colors in an image, that is, the proportion of different colors in the whole image, and it is especially suitable for describing those images that are difficult to automatically segment and do not need to consider the spatial position of objects. Its disadvantage is that it can't describe the local distribution of colors in the image and the spatial position of each color, that is, it can't describe a specific object or object in the image.

The most commonly used color space: RGB color space, HSV color space.

Color histogram feature matching methods: histogram intersection method, distance method, center distance method, reference color table method and cumulative color histogram method.

(2) Color setting

Color histogram method is a global color feature extraction and matching method, which can not distinguish local color information. A color set is an approximation of a color histogram. Firstly, the image is transformed from RGB color space to visually balanced color space (such as HSV space), and the color space is quantized into several handles. Then, the image is divided into several regions by automatic color segmentation technology, and each region is indexed by a certain color component of the quantized color space, thus representing the image as a binary color index set. In image matching, the distance between different image color sets and the spatial relationship of color regions are compared.

(3) Color Moment

The mathematical basis of this method is that any color distribution in an image can be represented by its moments. In addition, because the color distribution information is mainly concentrated on the low-order moments, only the first-order moments (mean), second-order moments (variance) and third-order moments (skewness) of colors are enough to represent the color distribution of an image.

(4) Color aggregation vector

The core idea is to divide the pixels belonging to each handle of the histogram into two parts. If the area of the continuous area occupied by some pixels in the handle is greater than a given threshold, the pixels in this area are regarded as aggregated pixels, otherwise they are regarded as non-aggregated pixels.

(5) Color correlation diagram

Double texture feature

(1) feature: Texture feature is also a global feature, which also describes the surface properties of the scene corresponding to the image or image area. However, because texture is only the surface feature of an object, it cannot fully reflect the essential attributes of the object, so it is impossible to obtain high-level image content only by using texture features. Different from color features, texture features are not based on pixel features, but need to be statistically calculated in an area containing multiple pixel points. In pattern matching, this regional feature has great advantages, and it will not be unable to match successfully because of local deviation. As a statistical feature, texture features often have rotation invariance and strong noise resistance. However, texture features also have their disadvantages. An obvious disadvantage is that when the resolution of the image changes, the calculated texture may have great deviation. In addition, because it may be affected by illumination and reflection, the texture reflected from the 2-D image is not necessarily the real texture on the surface of the 3-D object.

For example, reflection in water and mutual reflection on smooth metal surfaces will lead to changes in texture. Because these are not the characteristics of the object itself, sometimes these false textures will cause "misleading" when texture information is applied to retrieval.

It is an effective method to use texture features when retrieving texture images with great differences in thickness and density. However, when there is little difference between textures, such as thickness and density, it is difficult to accurately reflect the differences between textures with different visual feelings.

(2) Common feature extraction and matching methods.

Classification of texture feature description methods

The typical representative of (1) statistical method is a texture feature analysis method called gray * * * generation matrix. On the basis of studying various statistical characteristics of the * * generating matrix, Gotlieb and Kreyszig and others obtained four key characteristics of the grey * * generating matrix through experiments: energy, inertia, entropy and correlation. Another typical statistical method is to extract texture features from the autocorrelation function of the image (that is, the energy spectrum function of the image), that is, to extract texture parameters such as thickness and directionality by calculating the energy spectrum function of the image.

(2) Geometric method

The so-called geometric method is a texture feature analysis method based on the theory of texture primitives (basic texture elements) According to the theory of texture primitives, a complex texture can be formed by repeating several simple texture primitives in a certain regular form. Among geometric methods, there are two more influential algorithms: Voronio chessboard feature method and structure method.

(3) Model method

The model method is based on the image construction model, and the parameters of the model are used as texture features. Typical methods include random field model method, such as Markov random field (MRF) model method and Gibbs random field model method.

(4) signal processing method

Texture feature extraction and matching mainly include: gray * * * generation matrix, Tamura texture feature, autoregressive texture model, wavelet transform and so on.

The feature extraction and matching of gray * * generating matrix mainly depend on four parameters: energy, inertia, entropy and correlation. Tamura texture features are based on the psychological research of human visual perception texture, and six attributes are proposed, namely: roughness, contrast, direction, line image, regularity and roughness. Autoregressive Texture Model (SAR) is an application example of Markov Random Field (MRF) model.

Triform feature

(1) Features: All kinds of shape-based retrieval methods can effectively use the objects of interest in the image for retrieval, but they also have some common problems, including: ① At present, shape-based retrieval methods still lack a relatively perfect mathematical model; ② If the target is deformed, the retrieval results are often unreliable; (3) Many shape features only describe the local properties of the target, and it often requires high computing time and storage capacity to completely describe the target; ④ The shape information of the target reflected by many shape features is not completely consistent with human intuitive feelings, or the similarity of feature space is different from that perceived by human visual system. In addition, the three-dimensional object represented in the two-dimensional image is actually only the projection of the object on a certain plane in space, and the shape reflected from the two-dimensional image is often not the real shape of the three-dimensional object, which may cause various distortions due to the change of viewpoint.

(2) Common feature extraction and matching methods.

Ⅰ Several Typical Description Methods of Shape Features

Generally, there are two ways to express shape features, one is contour feature, and the other is region feature. The contour feature of the image is mainly aimed at the outer boundary of the object, while the regional feature of the image is related to the whole shape region.

Several typical description methods of shape features;

(1) boundary feature method This method obtains the shape parameters of the image by describing the boundary features. Among them, Hough transform method and boundary direction histogram method are classical methods to detect parallel lines. Hough transform is a method of connecting edge pixels to form a closed boundary by using the global characteristics of an image. Its basic idea is the duality of points and lines. The boundary direction histogram method first differentiates the image to get the image edge, and then histograms the size and direction of the edge. The usual method is to construct the image gray gradient direction matrix.

(2) Fourier shape descriptor method

The basic idea of Fourier shape descriptor is to use the Fourier transform of the object boundary as the shape description, and use the closure and periodicity of the regional boundary to transform the two-dimensional problem into a one-dimensional problem.

Three shape expressions are derived from boundary points, namely curvature function, centroid distance and complex coordinate function.

(3) Geometric parameter method

The expression and matching of shapes adopt simpler regional feature description methods, for example, the shape factor method is used for quantitative measurement of shapes (such as moments, areas, perimeters, etc.). ). In QBIC system, geometric parameters such as roundness, eccentricity, principal axis direction and algebraic moment invariants are used for image retrieval based on shape features.

It should be noted that the extraction of shape parameters must be based on image processing and image segmentation, and the accuracy of parameters will inevitably be affected by the segmentation effect. For the image with poor segmentation effect, even the shape parameters cannot be extracted.

(4) Shape invariant moment method

The moment of the area occupied by the target is used as the shape description parameter.

(5) Other methods

In recent years, the work in shape representation and matching also includes finite element method, steering function and wavelet descriptor.

Shape Feature Extraction and Matching Based on Wavelet and Relative Moment

In this method, a multi-scale edge image is obtained by using the modulus maxima of wavelet transform, and then seven invariant moments of each scale are calculated, which are then converted into 10 relative moments, and the relative moments of all scales are used as image feature vectors, thus unifying regions and closed and unsealed structures.

Four spatial relationship characteristics

(1) Features: The so-called spatial relationship refers to the mutual spatial position or relative direction relationship between multiple objects segmented in the image, and can also be divided into connection/adjacency relationship, overlapping/overlapping relationship and inclusion/inclusion relationship. Generally, spatial location information can be divided into two categories: relative spatial location information and absolute spatial location information. The former relationship emphasizes the relative position between targets, such as the relationship between up and down, left and right. The latter relationship emphasizes the distance and direction between targets. Obviously, relative spatial position can be deduced from absolute spatial position, but it is often very simple to express relative spatial position information.

The description and resolution of image content can be enhanced by using spatial relationship features, but spatial relationship features are often sensitive to the rotation, inversion and scale change of images or targets. In addition, in practical applications, it is often not enough to use only spatial information to express scene information effectively and accurately. In order to search, other features are needed besides spatial relationship features.

(2) Common feature extraction and matching methods.

There are two methods to extract the spatial relationship features of an image: one is to automatically segment the image and divide the objects or color areas contained in the image, and then extract the image features according to these areas and establish an index; Another method simply divides the image into several regular sub-blocks, and then extracts features from each image sub-block and establishes an index.