RU2483354C1 - Method for isolation of object in image - Google Patents

Abstract

FIELD: instrument making.

SUBSTANCE: according to the proposed method, two images of the object are shaped in series by means of an objective lens for further digital processing consisting in shaping of a differential image and its threshold processing. A parallel flat plate is installed at a convergent optical beam at shaping of the first focused images of objects and background in the objective lens of optical path. At formation of the second defocused image and background at a convergent optic beam there installed is a parallel flat plate with the thickness different from that of the first one. Differential image is shaped by subtraction of the second image from the first one.

EFFECT: providing maximum signal-noise ratio and excluding the influence of geometrical noises.

1 dwg

Description

The invention relates to optical instrumentation, the fields of photonics and computer engineering, mainly to infrared (IR) systems, which perform spatial image sampling using a multi-element matrix radiation detector used in various fields of science, technology, and the national economy, in particular, for the detection of foreign objects in given area of space.

There is a method of distinguishing objects, which consists in receiving radiation from objects and the background, forming their image in the plane of the radiation receiver and selecting objects according to the corresponding a priori distance to the object (see US patent No. 3221591, class NPK 350-10, 1972).

The disadvantages of this method include the low probability of distinguishing objects located at a finite distance from the device, with an a priori unknown remote background.

Also known is a method of extracting an object from an input image using a reference image (Japanese Patent IP 08-323299, publ. November 20, 1996, G06K 9/46). This invention relates to methods for image processing and is intended for accurate automatic determination of the area containing the desired object in the input image, and its selection. The threshold is determined by the data on the intensity of the boundaries of the objects obtained by subtracting the previous frame from the next. The designation of the main area occurs due to the analysis of the contour line obtained by subtraction, and further recognition occurs when comparing with the standard contained in the memory. The threshold is set as the number of pixels contained in the difference image. The definition of the area is performed by highlighting a certain part of the original image when combined with the main area.

The disadvantage of this method is the need to use a reference image, which makes it impossible to recognize objects that are not in the database (only their detection in the database is possible), and it also recognizes only the only type of object in the current frame, and it is necessary to take into account both chromatic and and spatial-geometric features of the object. The use of multi-stage algorithms for dividing the image into segments, required to achieve a high recognition probability, significantly increases the software complexity of the method.

The invention of Japan patent JP 2008-66775, publ. March 14, 2008, G06K 9/00, “Device for detecting an object, method of its control, program, information storage medium and device in which they are used” is characterized by the following sequence of actions:

- receiving an input image;

- selection of criteria by which the object will be detected;

- detection and recognition of an object in the frame according to the established criterion;

- output images with information of interest.

This invention is intended to solve the following disadvantages of the known solutions for the recognition of objects in the frame:

- extremely high load on the computing resources of the device, leading to the use of expensive electronics and high power consumption;

- the need for time separation between detection and recognition operations to optimize computing resources, which increases the time to obtain the final result.

The main objective of this invention is the quick output to the display device of the recognized object in the frame while processing the next object - the combination of the operational display of the results together with the parallel processing of the incoming data allows to increase the speed of the system without the use of additional computing resources.

Among the disadvantages of the method should be highlighted:

- the use of several detection criteria leads to a set of results from which you want to choose the best, which complicates the structure of the processing algorithm and significantly increases the amount of memory required to store the results;

- to increase the recognition efficiency, it is necessary to process a series of images, while the conditions can vary from frame to frame, which is also not optimal from the point of view of the time of obtaining results and the amount of memory used;

- the method is based solely on software correction of results and does not allow the use of hardware to adjust external conditions during measurements (i.e. there is no control of optics, position of the FPU, etc.).

As a prototype, the invention can be considered the “Method for extracting an image of an object from the background” (US patent No. 4783833, publ. 11/08/1988, class G06K 9/00), in which the first still image of the background is formed using the ECO lens as which can be an object, register it using a matrix FPU and electronically process it in order to isolate and enhance the brightness of its boundaries. Then, when a moving object appears, a second image is formed, recorded similarly to the first, and the brightness change of the pixels of the background image is recorded by subtracting the second registered image from the first. Due to the threshold processing, a pixel is selected according to a predetermined criterion, the signal value from which exceeds a predetermined threshold, which creates a contrast image of the boundaries of the object, which contributes to its confident selection.

To increase the accuracy of the selection of a moving object, the receiver module is shifted in the vertical and horizontal directions during the frame time, which is equivalent to setting a certain law for the displacement of the background image, while the object moves according to its own law. In this case, frame-by-frame comparison allows achieving the most accurate combination of images of a moving object in time, providing confident selection of the object in a dynamic environment.

Thus, the recognition of a moving object (selection of its contour) is carried out by subtracting the subsequent and previous frames, during which the receiving module is shifted in two planes perpendicular to the optical target axis of the device, using the displacement of objects of its frame area. Moreover, in the present invention, the problems are solved by displacing the object in depth (i.e., along the optical axis of the device) with its location in the frame unchanged.

The main disadvantage of this method is the rather low probability of recognition of objects due to the influence of the noise component due to the heterogeneity of the sensitive surface of the matrix FPU.

The task to which the invention is directed is suboptimal image filtering, close to that implemented in the human eye.

This problem is solved due to the fact that the claimed method of selecting an object in an image, which consists in the sequential formation by the lens of two images of objects and a background on the surface of the matrix FPU in the form of a matrix of values of electrical signals subjected to pixel-by-pixel recording and digital processing with subsequent formation of difference images by subtracting one registered image from another, the implementation of threshold processing of the differential image and determining the pixel, the brightness value and which exceeded the threshold, when forming the first focused image of objects and background in the objective of the optical path in a converging optical beam, a plane-parallel plate is installed, when forming the second defocused image and background in a converging optical beam, a plane-parallel plate is set with a thickness different from the thickness of the first, while the differential image form by subtracting the second image from the first.

The technical result is provided by the given set of features and is expressed by the achievement of the maximum ratio of the received output signal to the level of interference from clouds and noise, which is associated with the independence of the signal from the level of the dark elements of the matrix; The extraction of the optical signal occurs on the same element of the matrix, so it occurs accurately, without the influence of geometric noise.

The essence of the invention is illustrated in the drawing, which shows a schematic diagram that implements the algorithm for selecting images of objects on a remote background.

A device that implements the claimed method, contains a lens 1, a pair of different thickness non-coaxial plane-parallel optically transparent plates 2 and 3 with the same transmittance mounted on the axis of the rotation drive 4, parallel to the optical axis, a multi-site photodetector (MFP) 5, an electronic image processing path 6 and display unit 7. In this case, the path 6 includes an analog to digital converter (ADC) 8, memory devices 9, 9 ', a subtractor 10, a comparison unit with a pre-selected time Big 11 and a digital to analog converter (DAC) 12.

The principle of operation of a device that implements the claimed method is that the infrared radiation from the object and the background is projected by the optical system 1 of the optoelectronic device (OED) through a plane-parallel transparent plate 2 on the MFP 5. In this case, the first image is exposed by introducing onto the optical axis of the lens 1 in a converging beam, a transparent thin plate 2, the formed image is read pixel by pixel from the MFP 5 in the form of a matrix of values of the electric signal, digitized in ADC 8 and stored in device 9. Then m using a rotation drive 4, a transparent thick plane-parallel plate 3 is introduced into the converging beam onto the optical axis of the lens 1, thereby forming an image defocused by the lens 1 (not designed to work with a thick plate), similarly receive an electronic signal, reading it pixel by pixel from the MFP 5, digitized in the ADC 8 and stored in the device 9. To determine the type of used plate 2 or 3 for synchronization with the process of polling the matrix mounted marker sensor (not shown); when changing the rotating plates, the value stored in the device 9 'is compared with the value in the device 9 and the difference image obtained in block 10 is subjected to threshold processing in block 11, and a signal is extracted by fixing pixels whose brightness exceeds a selected threshold. Using well-developed and effective methods of processing digital images, they filter the signal from interference and convert it in the block of the DAC 12 into an analog form for presentation on the screen of the monitor 7 in the form of a signal convenient for the operator to perceive.

Claims (1)

A method of isolating an object in an image, which consists in sequentially generating two images of objects and a background on the surface of a matrix FPU in the form of a matrix of values of electrical signals subjected to pixel-by-pixel recording and digital processing, followed by the formation of differential images by subtracting one registered image from another, and performing threshold differential processing image and determining a pixel whose brightness value exceeded a threshold, characterized in that and the formation of the first focused image of objects and the background in the objective of the optical path in the converging optical beam set plane-parallel plate, when forming the second defocused image and the background in the converging optical beam set plane-parallel plate thickness other than the thickness of the first, while the difference image is formed by subtracting the second image from first.