CN104181176B - The method that liquid article automatic identification is carried out in X-ray wraps up image - Google Patents

Abstract

The invention discloses a method for automatic identification of liquid items in an X-ray package image. A method based on container symmetry and loosely combined edges is used to identify whether there is a shape of liquid items in the image. Based on the edge points of the container, the point pairs of edge points are obtained step by step, and then the container segment is formed by the point pair, that is, the main body or a part of the container, and finally the container is combined to form the whole container; if the results of the above steps, If all meet the characteristics of the container, it is considered that a liquid item has been found; the advantage of the present invention is that it can automatically identify the liquid item with a specified volume in the package, has a good balance between the detection rate and the false alarm rate, and ensures that most of the package will be While liquid items are detected, the false alarm rate can be significantly reduced.

Description

Method for Automatic Recognition of Liquid Items in X-ray Parcel Images

technical field

The invention relates to a method for automatic identification of liquid articles in X-ray package images, and belongs to the technical field of X-ray safety inspection.

Background technique

At present, in the civil aviation security inspection operation specification, liquids of more than 100 milliliters are prohibited from being carried. In the future, with the development of liquid detection technology, it is expected that the mode of liquid take-out inspection will be implemented, that is, passengers need to take out the liquid items carried in the luggage package, and the liquid items will be scanned and interpreted by the security inspection equipment alone. Ultimately, the model that the market expects is to allow liquids to be automatically detected in packages to determine their danger. In the above three modes, a technology that can automatically identify the liquid items in the package is needed. In the first two modes, it is necessary to prevent the liquid items from being hidden in the package and pass through the security check. The last mode needs to be locked. The position of the liquid object is then detected in a targeted manner.

At present, the most prominent image recognition work in the field of X-ray security inspection technology is Canada's OptoSecurity, which has applied for a series of patents on object recognition in package images, and has established strategic partnerships with leading companies such as Heyman. partnership. In its US patent 20120275646A1, a method for automatic identification of liquid items in X-ray package images is mentioned. This method firstly segments the area with liquid material characteristics in the image, extracts the material, grayscale, and shape features of the relevant area, and identifies whether there is a liquid object by fusing the segmented areas. In addition, the papers on the automatic detection of liquid objects in X-ray images seen in other individual documents are all attempts in the exploratory stage.

Based on the identification of segmented regions, liquid items with relatively simple backgrounds can be identified. However, when liquid objects overlap with more complex background objects, the identification and fusion of the above-mentioned segmented regions will become very difficult; if the requirements for the prominence of the segmented regions are lowered, there will be too many searchable segmented regions, This results in more false alarms.

At present, in the field of security inspection, liquid items are prohibited from being carried on the plane. Usually, passengers are prevented from putting liquids in the luggage by relying on the self-consciousness of passengers and the visual inspection of security images by security inspectors. If the liquid items in the package can be effectively and automatically identified, the work intensity of the security inspector can be reduced and the safety factor can be improved. At the same time, the technology of automatic identification of liquid items also lays a solid foundation for the technology of automatic detection of liquids in the package in the future, that is, in the aforementioned third inspection mode where the liquid does not need to be taken out, the location of the liquid is first determined.

Contents of the invention

The purpose of the present invention is to provide a method for automatic identification of liquid items in X-ray package images that can overcome the above technical problems. The present invention uses a method based on container symmetry and loosely combined edges to identify Whether there is a form of liquid items, which can enhance the stability of the recognition algorithm.

The present invention is based on the edge points in the image, step by step upwards, first obtain the point pairs of the edge points, and then form the container segment through the point pairs, that is, the main body of the container or a part of the segment, and finally form the whole container by combining the container segments; if If the results of the above steps all conform to the characteristics of the container, it is considered that a liquid item has been found.

After obtaining the security inspection image of the package, the method of the present invention specifically includes the following steps:

1) Image resolution reduction;

2) Image edge point detection;

3) Search and form edge point pairs that are symmetrical to each other;

4) Hough transform of point pairs, so as to find the location parameters where point pairs are concentrated and containers may appear;

5) According to the position parameter found in step 4), look for the container segment;

6) Find the container characteristics of each container segment, including grayscale distribution, container wall material, bottleneck or cap shape;

7) Find the liquid item in the image by combining suitable container segments together to form the body of the container.

The method of the present invention is a strategy based on edge information to form the container body from microscopic edge points, and the present invention utilizes the constraints of the container's unique shape, gray scale, and material properties.

The invention has the advantage of being able to automatically identify liquid items of a specified volume in a package, having a good balance between detection rate and false alarm rate, and can significantly reduce the false alarm rate while ensuring that most of the liquid items in the package are detected .

Description of drawings

Fig. 1 is a schematic diagram of the principle of the present invention;

Fig. 2 is a schematic diagram of the symmetric relation of the point pairs of the edge points of the present invention, and the line segment with the arrow represents the gradient direction of the edge points;

Fig. 3 is the schematic diagram of the X-ray attenuation curve of each pixel point on the connection line of a point pair of the present invention;

Fig. 4 is the schematic diagram that the present invention searches the point pair that forms bottleneck and bottle cap contour at one end of container segment;

Fig. 5 is a schematic diagram of combining the container sections of the present invention into a container.

detailed description

The present invention will be described in detail below in conjunction with the accompanying drawings and embodiments. Fig. 1 has provided the schematic diagram of the principle of the method of the present invention, represents from edge point, to point pair, to container segment, finally forms the recognition process of container, and this principle helps to overcome liquid because of overlapping imaging with other objects and make bottle Issue with contours being truncated into multiple segments.

The specific processing steps are as follows:

1. Reduce the length and width of the X-ray wrapped image to 1/4 of the original through image down-sampling technology, and use median filtering to denoise, so as to avoid excessive reaction to tiny edges in subsequent steps;

2. Detection and grouping of edge points;

The detection of edge points uses the Canny algorithm, and then according to the material value M of each edge point detected, all edge points are grouped into: 1) metal points, 2) glass or ceramic points, 3) approximate plastic points; according to the edge point Divide all edge points into 9 groups, equally divide the angle space from 0 to 180 degrees; combine the above two aspects, divide the edge points of the image into 27 groups in total The superscript m stands for metal, g stands for glass, and p stands for plastic.

3. Search for edge point pairs, note that the points described below are the edge points extracted in step 2;

Suppose an edge point q belongs to the plastic point and belongs to group, then only the In these three groups, search for the edge point q' that can match q, and form a point pair. For metal points or glass points, similarly search for point pairs in groups of the same material; in the search, exclude the gradient direction of two points The combination of the difference exceeding the threshold and the gradient value difference exceeding the threshold. The meaning of the point pair is shown in Figure 2; for the point pair q, q', let L be the connecting line between q and q', and let the straight line where the respective gradient directions of q, q' are located be L ^q , L ^q' . The angles α(q,L) and α(q',L) between L and L ^q , L ^q' are approximately equal, and the gradient directions of q and q' are relative rather than opposite or the same towards.

4. Hough transform of point pairs;

First, for the collected point pair (q, q'), find the symmetric axis line S of this point pair, S is a straight line passing through the midpoint of the line L connecting q, q' and the direction is orthogonal to L. According to the polar coordinate expression S(θ, r) of the straight line S of the symmetry axis and the distance d between two points, the current point pair is accumulated to a Hough( θ, r, d) in the 3-dimensional table, in which the 3-dimensional components each select an appropriate table span; θ represents the slope angle of the straight line S, and r represents the vertical distance from S to the origin of the image. After performing the above operations on all point pairs, find the parameter combination (θ, r, d) with the cumulative score exceeding the threshold and the local maximum in the Hough (θ, r, d) table as the object of the next analysis.

5. Find the container segment;

The container segment represents a section of the container area, in which the contours of both sides of the container appear symmetrically; since objects in X-ray images are often superimposed and imaged, it is often impossible to find the whole container immediately, so the container segment is first searched. For the (θ,r,d) combination that meets the requirements in step 4, vertically project the two points of all point pairs belonging to the parameter combination onto the straight line S(θ,r), and investigate on S(θ,r) The distribution of these projected points is to find the line segment with continuous projected points on S(θ, r) and the local density of projected points is high enough. On S(θ, r), the line segments of such projected points appear continuously and with high density, forming the symmetry axis of a container segment, that is, taking the straight line S(θ, r) as the symmetry axis line, and appearing in this area A container segment with a width d, of course, is currently only a possible container segment; each container segment records its own symmetry axis line and the positions of the two ends of the symmetry axis segment. In addition, it is also necessary to perform local optimization on the point pairs under the jurisdiction of the container segment, and between the points of the point pairs with similar projected positions on S(θ, r), according to the point pair symmetry rule described in step 3, Find the pairwise correspondence of these points again, so as to form the truly optimal combination of point pairs locally relative to S(θ, r).

6. Find the container characteristics of the container segment;

The characteristics of the container mainly include: (1) whether there is a bulge of X-ray attenuation in the central area close to the symmetry axis in the container section; (2) judging the material of the container; (3) whether there is a bottleneck or a bottle cap at one end of the container section shape. The first two points (1) and (2) reflect the overall law of the projected gray scale of a liquid container, and the point (3) reflects the most typical parts of the container.

1) For the bulge or bulge in the middle of the container segment, as shown in Figure 3, establish the curve of the X-ray attenuation of each pixel on the line connecting all points to (q, q') of this container segment, in each curve , to investigate the comparison between the average attenuation of the 5 parts of the 1/5 on both sides, that is, the A and E regions, the 1/5 near the middle, that is, the B and D regions, and the 1/5 in the middle, that is, the C region. Here, the area number is directly used to represent the average gray level of the area. If C/A, C/B, C/D, C/E, B/A, and D/E are greater than the respective specified thresholds, it is considered to be attenuated here There is a bulge on the volume curve; if there is a bulge in the attenuation curve between most point pairs to which a container segment belongs, it is considered that this container segment has a bulge.

2) The identification of the glass and metal materials of the container, in the vicinity of the two points of each point pair of the container segment, if the unique material properties of the glass container or the material properties of the metal container are generally present, it is considered that the current container segment may correspond to glass Containers or metal containers, such as tin beverage cans. The material properties of the X-ray image can be represented by the combination of the material value M and the pseudo-color R value. Which material properties correspond to glass and metal, this law is statistically obtained in the training images prepared in advance, and a corresponding look-up table mechanism is established; other The container segment can be considered to correspond to a plastic container.

3) The search for the shape of the neck or cap, at the two ends of the container segment, extends the search for the neck or cap outside the container segment, which is mainly for small-mouth bottles; because the edge point detection of the neck or cap contour is often not Complete and asymmetric, so along the straight line S of the symmetry axis of the container segment, extend outwards at both ends of the container segment to search for edge points and form point pairs. The symmetric axis line of these point pairs is close to S; the specific method is: from the container Starting from the endpoint at one end of the segment, and moving along S to the direction outside the container segment by one unit distance, for example, the length of 1 pixel, the following search is performed at the current position S _p of S: at S _p along the direction of S The orthogonal straight line explores the edge points on both sides of S, and if one edge point is found within the appropriate range on both sides, check whether the two edge points can be determined as symmetric to S according to the principle described in step 3 The point is correct, but the threshold requirements in step 3 are reduced accordingly, and the material properties at the two edge points are required to be similar to the container material of the container segment. As shown in Figure 4, these newly searched edge point pairs are considered to possibly form the outline of the bottle neck or bottle cap of the container, and the variation trend of the spacing of these point pairs is analyzed whether it conforms to the container rules, so as to determine whether there is a bottleneck or a bottle cap:

a) According to the order in which the point pairs are found, the distance between the point pairs is required to be from large to small, and the curve formed by the distance between each point pair

In the line, the second order derivative crosses 0 at no more than 2 places;

b) In the last part of this group of point pairs, the spacing width of the point pairs is relatively stable, corresponding to the shape of the bottle cap;

c) The ratio of bottleneck length to container segment width should be reasonable.

7. Combined container segment;

To sum up, the container segment is mainly divided into the following types; the glass container segment is called the T0 type, and the other container segment types are listed in Table 1. If the value of the first column in Table 1 is No, it means this row The container segment type corresponds to plastic containers.

Table 1 Types of container segments

Whether it is a metal container the protruding has a neck or cap type number Yes Yes Yes T1 Yes Yes no T2 Yes no Yes T3 Yes no no T4 no Yes Yes T5 no Yes no T6 no no Yes T7 no no no T8

Next, start to combine the container segments, as shown in Figure 5, combine the container segments with similar linear parameters of the symmetry axes, close distances to each other, and similar widths to form a container; Note: 1) A single container segment has It may correspond to 1 container; 2) When combining container segments, if the symmetry axes of the combined container segments are distributed in a serpentine shape, the current combination is canceled because in the X-ray security inspection image, the container projection may be curved , but it can only bend in one direction and will not form a serpentine bend.

At the same time, in the combination of container segments, set the rules according to common sense, for example, several T2 type container segments may be combined to form a metal soda can liquid item; or a T3 type container segment and several T4 type container segments form a 1 A flat metal container with a bottleneck; 1 T5 container segment and several T6 container segments form a plastic container, such as a typical bottled purified water; this is the case for a T7 container segment and several T8 container segments It is similar to plastic containers for shampoo; and the separate T8 type container segment is generally not considered to have containers. In short, the combination of container segments is carried out according to the form of common liquid items in life.

After finding a reasonable combination of container segments, it is necessary to examine the rationality of the size and aspect ratio of the container formed by the combination. For example, for a combination that is expected to be a can, if its width is too small or too large, or its length is too long or too short, this combination of containers can be excluded. Finally, for the remaining combination of containers, the upper layer software system is notified, and an alarm is given on the position of the corresponding container on the screen of the security inspection device.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the disclosure of the present invention are all It should be covered within the protection scope of the claims of the present invention.

Claims (1)

1. the method that liquid article automatic identification is carried out in X-ray wraps up image, it is characterised in that：With the marginal point in image Based on, step by step upwards, the point for obtaining marginal point first is right, and then the main body or by point to composition container section, i.e. container Partial Fragment, is finally combined the entirety for forming container by container section；If the result of each step, meets the spy of container above Point, then it is assumed that find 1 liquid article；After the safety check image for obtaining parcel, following steps are specifically included：

1) image resolution decreasing；

2) image border point detection；

3) search for and constitute；

4) point to Hough transformation, so as to find a little to concentration, the location parameter that is likely to occur of container；

5) according to step 4) location parameter that finds, find container section；

6) container characteristics of each container section, including the shape of intensity profile, chamber wall material, bottleneck or bottle cap are found；

7) suitable container section is combined, forms the main body of container, so as to find the liquid article in image；

The method that liquid article automatic identification is carried out in X-ray wraps up image is that one kind make use of the distinctive shape of container Shape, gray scale, the constraint of material behavior based on marginal information, finally combination forms the strategy of container body from microcosmic marginal point Method.