CN101567080B - Method for strengthening infrared focal plane array image - Google Patents

Abstract

The invention discloses a method for strengthening an infrared focal plane array image, pertains to the field of infrared focal plane detector, particularly relates to the hardware realization of a specified image processing algorithm, and aims at extending image gray value and increasing computerization speed under the condition of limited FPGA storage resource. The method comprises the steps of median filtering, sub-point access and sub-section extension. The median filter adopted by the method eliminates spiced salt noise in the image and the sub-section extension effectively strengthens the detailed part of the dark image area. The algorithm requires no external memory, has low requirement for the FPGA storage resource and good real-time property and can effectively fit the high requirement of the infrared focal plane detector to image strengthening processing.

Description

A kind of method for strengthening infrared focal plane array image

Technical field

The invention belongs to the infrared focal plane detector field, be specifically related to a kind of method for strengthening infrared focal plane array image.

Background technology

Infrared imaging system develops along with the development of infrared eye.In first generation infrared imaging system, adopt detector array, be embodied as picture by the one dimension optical mechaical scanning.Along with CCD (ChargeCoupled Device, charge coupled device) maturation of correlation technique, to 20th century the mid-1970s, the appearance of IRFPA (Infrared Focal Plane Array, infrared focal plane array) detector indicates second generation infrared imaging system---the birth of staring infrared imaging system.Compare with detector array, focal plane detector imaging has spatial resolution height, strong, the frame frequency advantages of higher of detectivity, becomes the main flow device of infrared imagery technique just rapidly.Staring infrared imaging system has begun to be widely used in civil areas such as night vision, sea rescue search, astronomy, industrial hot-probing and medical science at present, is the developing direction of infrared imaging system.Yet because aspects such as manufactured materials, technology and working environments, the image ubiquity target contrast of infrared focal plane array output a little less than, shortcoming such as background detail is fuzzy.Nominal output data width as certain infrared focal plane array is 16, yet in the actual use, no matter what kind of scene is, most of pixel value concentrates on 0x7EC0 on the scope of 0x82C0, and display effect is self-evident.So the image of infrared focal plane array output is carried out the pre-service of figure image intensifying, make it to be fit to application-specific, then very necessary.The method of figure image intensifying is divided into spatial domain method and frequency domain method two classes, and spatial domain method mainly is that each pixel in the image is operated; And frequency domain method is in certain transform domain of image, and image is operated, and revises the coefficient after the conversion, and then carries out the image after inverse transformation obtains handling.

Frequency domain method is regarded image as a kind of 2D signal, and it is carried out the conversion of time domain to frequency domain.Adopt low-pass filtering (promptly only allowing low frequency signal pass through) method, can remove the noise among the figure; Adopt high-pass filtering method, then can strengthen high frequency signals such as edge, make fuzzy picture become clear.But with dsp processor piece image is carried out the positive inverse transformation of frequency domain, calculated amount is big, is difficult to be fit to the demanding occasion of real-time.Frequency domain method strengthens image at frequency domain, and no matter high-pass filtering or low-pass filtering all can be destroyed original image significantly, and a lot of effectively information all might be by filtering.The frequency domain method calculation of complex is difficult to realize so frequency domain method is applied to the high real-time occasion its limitation being arranged by FPGA.

Histogram equalization is the image enchancing method in a kind of typical space territory, and the central idea that histogram equalization is handled is from becoming the even distribution in whole tonal ranges between certain gray area of relatively concentrating the grey level histogram of original image.Original image gray-scale value r normalization is between 0～1, and p (r) is the probability density function of original image intensity profile.In fact histogram equalization is handled is exactly to seek a greyscale transformation function T, gray-scale value s=T (r) behind feasible the variation, wherein, s is normalized to 0～1, promptly set up the mapping relations between r and the s, require to handle the probability density function p (s)=1 that the back gradation of image distributes, expect that all gray level probabilities of occurrence are identical.

For the digital picture discrete case, the calculation procedure that its histogram equalization is handled is as follows:

(1) histogram of statistics original image

$p_r\left(r_k\right)=\frac{n_k}{n}$

In the formula, r _kBe normalized input picture gray scale, n _kBe that Normalized Grey Level equals r in the input picture _kNumber of pixels, n is the sum of all pixels of input picture.

(2) compute histograms cumulative distribution curve

$S_k=T\left(r_k\right)=\underset{j=0}{\overset{k}{\mathrm{\Σ}}}{p}_r\left(r_j\right)=\underset{j=0}{\overset{k}{\mathrm{\Σ}}}\frac{n_j}{n}$

(3) carry out image gray-scale transformation with cumulative distribution function as transforming function transformation function

According to the cumulative distribution function that calculates, set up the corresponding relation between input picture and the output image gray scale, the gray scale after the conversion to be reverted to original scope at last.

Histogram equalization is applied in a lot of fields as a kind of image processing method of basis, but be to realize mostly by DSP or CPU programming, its advantage is that dirigibility is than higher, debugging is convenient, maximum shortcoming is to be difficult to accomplish in real time or quasi real time processing that this is unacceptable in some field.And use FPGA to realize to solve the difficult problem of real-time processing well.But doing statistics with histogram needs size to be about the storage space of a view picture figure usually, " FPGA of video image grey scale signal histogram equalization realizes " literary composition as " application of electronic technology " o. 11th in 2006, advantage is the effect that histogram equalization can play the figure image intensifying well, weak point is that itself and unresolved statistics with histogram need expend the problem of a large amount of storage resources, just having expanded a slice SDRAM in the FPGA outside simply does statistics with histogram, increase cost, reduced the integrated level of system.

In " microcomputer information " 2007 the 63rd volume 6-2 phases " based on the linear enhancement algorithms of the real-time infrared image of FPGA " literary composition, utilize 5 two field pictures, carry out the method for interframe iteration, obtain maximal value and minimum value in the image, carry out one section linear stretch.Its advantage is need not the buffer memory entire image, storage resources is required low, but the efficiency of algorithm of its interframe iteration is lower, can only obtain maximal value and minimum value in the image, the stretching dirigibility is very low.

Summary of the invention

The object of the present invention is to provide a kind of method for strengthening infrared focal plane array image, this method can strengthen the detail section in territory, dark picture areas effectively under limited FPGA resources supplIes, improved arithmetic speed.

Method for strengthening infrared focal plane array image provided by the invention, its step comprises:

The 1st step medium filtering

The n two field picture is carried out medium filtering, and n is the frame number of pending image;

The 2nd step waypoint obtains

Utilize following formula (I), (II) and formula (III) to calculate the minimum gradation value X of the n two field picture of process medium filtering respectively _Min ⁿ, maximum gradation value X _Max ⁿGray-scale value X with waypoint _b ⁿ

$\left\{\begin{array}{c}{X}_{\min }^n=X_{\min }^{n-1}+{\mathrm{Step}}_{\min }^n\\ {\mathrm{Step}}_{\min }^n=K_{\min }^n*\mathrm{\ΔX}\\ K_{\min }^n=M*N*\min \%-\min \_{\mathrm{counter}}^n\end{array}\right.$ Formula (I)

$\left\{\begin{array}{c}{X}_{\max }^n=X_{\max }^{n-1}+{\mathrm{Step}}_{\max }^n\\ {\mathrm{Step}}_{\max }^n=K_{\max }^n*\mathrm{\ΔX}\\ K_{\max }^n=M*N*\max \%-\max \_{\mathrm{counter}}^n\end{array}\right.$ Formula (II)

$\left\{\begin{array}{c}{X}_b^n=X_b^{n-1}+{\mathrm{Step}}_b^n\\ {\mathrm{Step}}_b^n=K_b^n*\mathrm{\ΔX}\\ K_b^n=M*N*b\%-b\_{\mathrm{counter}}^n\end{array}\right.$ Formula (III)

X _Min ^N-1, X _Max ^N-1And X _b ^N-1Be respectively the gray-scale value of minimum gradation value, maximum gradation value and the waypoint of n-1 two field picture, M*N is the resolution of n two field picture, min%, max% and b% represent user's preset proportion value respectively, represent respectively in the real image that less than the minimum gradation value in the stretching algorithm, the pixel of the gray-scale value of maximum gradation value and waypoint accounts for the ratio of total pixel; Min_counter ⁿ, max_counter ⁿAnd b_counter ⁿBe respectively in the n two field picture gray-scale value less than X _Min ^N-1, X _Max ^N-1And X _b ^N-1Number of pixels, Δ X is the iteration weights;

The 3rd step segmentation stretches

The X that utilizes the 2nd step iteration to obtain _Min ⁿ, X _Max ⁿAnd X _b ⁿThe gray-scale value X that the n+1 two field picture that stretches, segmentation stretch and export behind the n+1 two field picture _Out ^N+1:

$X_{\mathrm{out}}^{n+1}=\left\{\begin{array}{cc}0& \&ForAll;X_{\mathrm{in}}^{n+1}<X_{\min }^n\\ S1*(X_{\mathrm{in}}^{n+1}-X_{\min }^n)/(X_b^n-X_{\min }^n)& \&ForAll;X_{\min }^n<X_{\mathrm{in}}^{n+1}<X_b^n\\ S2*(X_{\mathrm{in}}^{n+1}-X_b^n)/(X_{\max }^n-X_b^n)& \&ForAll;X_b^n<X_{\mathrm{in}}^{n+1}<X_{\max }^n\\ S1+S2& \&ForAll;X_{\mathrm{in}}^{n+1}>X_{\max }^n\end{array}\right.$

X wherein _In ^N+1Be the original gray scale of n+1 frame input picture, S1 is for distributing to [X _Min ⁿ, X _b ⁿ] gray level, S2 is for distributing to [X _b ⁿ, X _Max ⁿ] gray level, S1, S2 is set by the user, the gray level that S1+S2 need be stretched to for the user.

The preferred Fast Median Filtering algorithm of above-mentioned medium filtering process, this algorithm at first sort to three column elements of a 3*3 window; Three row elements for the window after the row ordering sort then; Ask the intermediate value of three elements on the diagonal line at last, the gained intermediate value is exactly the intermediate value of 9 elements.

The present invention is divided into two sections of bright object and dark areas with image on the basis of background technology, stretch respectively, has strengthened the detail section of image large tracts of land dark areas effectively, has suppressed the influence of the bright object of small size to the stretching algorithm effect.Develop interframe iterative algorithm, stably obtain every section waypoint rapidly based on negative feedback thought.With infrared focal plane array output be w bit digital image quantization to the k position, make it to be easy to processing.W=16 or 14 generally, the k value is selected by the user, is generally 8.This algorithm can independently be realized by FPGA fully, need not external memory storage and the cooperation of DSP.Particularly, the inventive method has following technical characterstic:

(1) institute of the inventive method all need not the buffer memory entire image in steps, and the present invention greatly reduces the requirement of algorithm to storage resources like this.Because the reason of IC manufacturing process, built-in high capacity storage resources can improve the manufacturing cost of chip in the chip.Low storage consumption characteristics of the present invention make it can be used as an IP kernel and are integrated in a SOC system with lower cost.If in FPGA, realize the present invention, low in resources consumption, greatly reduced the realization cost, improved the integrated level of system.

(2) in the waypoint obtaining step, need not cache image, in the image input, can finish iterative operation.With X _Min ⁿBe retrieved as example, iteration step length V ⁿChoose based on negative feedback thought, step-length is variable, step-length was bigger when iteration began, and worked as X _Min ⁿAfter moving closer to actual value, step-length diminishes gradually, thereby reaches higher iteration precision, after general 10 two field pictures, and X _Min ⁿ, X _Max ⁿ, X _b ⁿCan tend towards stability.X _Min ⁿ, X _Max ⁿ, X _b ⁿIterative operation make full use of the characteristics of FPGA concurrent operation, carry out simultaneously, improved the efficient of algorithm operation greatly.

(3) segmentation stretches and image stretch can be arrived any gray level.Infrared focal plane array image output at present mostly is 16, and for most of target detection track algorithms, only needs the gray-scale map of 8 precision to get final product.

(4) in the segmentation stretching step, consider that infrared focus plane salt-pepper noise situation is comparatively under the condition of severe, medium filtering may be with all salt-pepper noise filterings, and segmentation stretches and chooses the gray scale maximal value X of the gray-scale value at histogram two ends min% and max% place as image _MaxWith minimum gray value X _Min, can obtain the maximal value and the minimum value of stable image like this, for algorithm for image enhancement provides stable parameter.General min gets 1～3, and max gets 97～99.

(5) in the segmentation stretching step, the method of adopt dividing two sections stretchings is divided into two sections of bright object and dark areas with image, stretches respectively, strengthened the detail section of image large tracts of land dark areas effectively, suppressed of the influence of the bright object of small size the stretching algorithm effect.

The preferred Fast Median Filtering algorithm of the present invention only needs three clock period just can try to achieve intermediate value, and very large raising has been arranged on speed.This algorithm does not reduce the ordering number of times, and preceding two clock period, 3 three input sequencing devices all will carry out three times ordering, and finally obtaining intermediate value needs 21 minor sorts.But carrying out relatively simultaneously of preceding two clock period 3 three input sequencing devices.This algorithm adopts the method for area throw-over degree, excavated the parallel ability of FPGA to greatest extent, so its lifting aspect speed also is maximum.

Description of drawings

Fig. 1 is the process flow diagram of the inventive method;

Fig. 2 is square window and median filtering algorithm process flow diagram;

Fig. 3 is a waypoint obtaining step status transition chart;

Fig. 4 is a thermal imaging system output digital image sequential;

Fig. 5 is an original graph;

Fig. 6 is the image after the enhancement process.

Embodiment

The present invention is divided into two sections of bright object and dark areas with image, stretches respectively, has strengthened the detail section of image large tracts of land dark areas effectively, has suppressed the influence of the bright object of small size to the stretching algorithm effect.

Key of the present invention is to obtain three gray-scale values of n two field picture, i.e. the minimum gradation value X of this two field picture _Min ⁿ, maximum gradation value X _Max ⁿGray-scale value X with waypoint _b ⁿ, being used to the n+1 two field picture that stretches, n is the frame number of image, establishes the image resolution ratio size and is M*N.

Gray-scale value is less than X _Min ⁿNumber of pixels account for the min% of total number of image pixels, promptly M*N*min%;

Gray-scale value is less than X _Max ⁿNumber of pixels account for the max% of total number of image pixels, promptly M*N*max%;

Gray-scale value is less than X _b ⁿNumber of pixels account for the b% of total number of image pixels, promptly M*N*b%;

The value of min and max and b is preestablished by the user, and generally speaking, the span of min is 1～3, and max is 97～99, and b sets according to the estimation ratio that bright object accounts for image area.Gray-scale value is less than X like this _b ⁿPixel promptly think large-area dark areas, distribute more gray level; Gray-scale value is greater than X _b ⁿPixel promptly think the bright object of small size to distribute less gray level.

Describe step of the present invention below in detail:

Method for strengthening infrared focal plane array image provided by the invention comprises medium filtering, and waypoint obtains, and segmentation stretches.

(1) medium filtering

Medium filtering preferably following manner carries out: the 3*3 square window is used mid{I1 as shown in Figure 2, I2, and I3, I4, I5, I6, I7, I8, I9} replace the original pixel value in former square window centre position.Afterwards, after a pixel and neighborhood territory pixel thereof disposed through medium filtering, the 3*3 square window will constantly move to right or enter a new line, up to all pixels in the data array of a width of cloth gray level image are all handled.

Median filtering algorithm in the square window: at first three column elements to a 3*3 window sort; Three row elements for the window after the row ordering sort then; Ask the intermediate value of three elements on the diagonal line at last, the gained intermediate value is exactly the intermediate value of 9 elements.

(2) waypoint obtains

X _Min ⁿ, X _Max ⁿAnd X _b ⁿ(waypoint) obtain the method for iteration frame by frame that adopts.With X _Min ⁿBe retrieved as example, X _Min ⁰Be X _Min ⁿInitial value, the nominal output data width of infrared focal plane array is 16, however in the actual use, no matter what kind of scene is, most of pixel value concentrates on 0x7EC0 on the scope of 0x82C0, so order $X_{\min }^0=0x7\mathrm{EC}0;$

Register min_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _Min ^N-1Number of pixels, Δ X is the iteration weights, Δ X is big more, and iteration speed is high more, but the precision step-down, balance iteration precision and iteration efficient generally can be made as 1.

Utilize following formula loop iteration,

$X_{\min }^n=X_{\min }^{n-1}+{\mathrm{Step}}_{\min }^n$

${\mathrm{Step}}_{\min }^n=K_{\min }^n*\mathrm{\&Delta;X}$

$K_{\min }^n=M*N*\min \%-\min \_{\mathrm{counter}}^n$

Register max_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _Max ^N-1Number of pixels,

Register b_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _b ^N-1Number of pixels,

X _Max ⁿAnd X _b ⁿObtain with identical method.After general 10 two field pictures, X _Min ⁿ, X _Max ⁿAnd X _b ⁿPromptly tend towards stability.

(3) segmentation stretches

Two sections drawing process are established wherein X _In ^N+1Be the original gray scale of n+1 frame input picture, X _Out ^N+1Be the gray scale that segmentation stretches and exports behind the n+1 two field picture, S1 is for distributing to [X _Min ⁿ, X _b ⁿ] gray level, S2 is for distributing to [X _b ⁿ, X _Max ⁿ] gray level.In the present embodiment image stretch to 8 gray level with 16, so S1+S2=255; Get S1=200, S2=55;

The X that n frame iteration obtains before utilizing _Min ⁿ, X _Max ⁿAnd X _b ⁿThe n+1 two field picture stretches.Transformation for mula is:

$X_{\mathrm{out}}^{n+1}=\left\{\begin{array}{cc}0& \&ForAll;X_{\mathrm{in}}^{n+1}<X_{\min }^n\\ S1*(X_{\mathrm{in}}^{n+1}-X_{\min }^n)/(X_b^n-X_{\min }^n)& \&ForAll;X_{\min }^n<X_{\mathrm{in}}^{n+1}<X_b^n\\ S2*(X_{\mathrm{in}}^{n+1}-X_b^n)/(X_{\max }^n-X_b^n)& \&ForAll;X_b^n<X_{\mathrm{in}}^{n+1}<X_{\max }^n\\ S1+S2& \&ForAll;X_{\mathrm{in}}^{n+1}>X_{\max }^n\end{array}\right.$

Below by by embodiment the present invention being described in further detail, but following examples only are illustrative, and protection scope of the present invention is not subjected to the restriction of these embodiment.

The present invention handles certain model refrigeration mode focal plane image of output in real time, and the image form size is: 320*256, frame frequency were 50 frame/seconds, and the pixel data bit wide is 16, and scene is the outdoor architecture thing, and focal plane output image sequential as shown in Figure 4.

Describe step of the present invention below in detail:

(1) medium filtering

In FPGA, generate two dual port RAMs, DPRAM0 and DPRAM1.Each dual port RAM is used for buffer memory 3 line data, and its size just can get final product by buffer memory 3 line data, adopts the method for ping-pong operation, when a dual port RAM is done medium filtering, and the view data of another dual port RAM storage focal plane output.The flow process of median filtering algorithm as shown in Figure 2, at first to the ordering of three column elements of a 3*3 window; Three row elements for the window after the row ordering sort then; Ask the intermediate value of three elements on the diagonal line at last, the gained intermediate value is exactly the intermediate value of 9 elements.After image enters this module, begin to export median-filtered result through behind 320+3 pixel clock.With frame useful signal 320+3 pixel clock cycle of time-delay, the medium filtering module promptly can focal plane sequential output median-filtered result figure.

(2) waypoint obtains

X _Min ⁿ, X _Max ⁿAnd X _b ⁿ(waypoint) obtain the method for iteration frame by frame that adopts.With X _Min ⁿBe retrieved as example, X _Min ⁰Be X _Min ⁿInitial value, register min_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _Min ^N-1Number of pixels, Δ X is the iteration weights.This example is got min=1, max=99, b=80, Δ X=1;

Utilize following formula loop iteration,

$X_{\min }^n=X_{\min }^{n-1}+{\mathrm{Step}}_{\min }^n$

${\mathrm{Step}}_{\min }^n=K_{\min }^n*\mathrm{\&Delta;X}$

$K_{\min }^n=M*N*\min \%-\min \_{\mathrm{counter}}^n$

Register max_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _Max ^N-1Number of pixels,

Register b_counter ⁿIn the middle record n two field picture, gray-scale value is less than X _b ^N-1Number of pixels,

X _Max ⁿAnd X _b ⁿObtain with identical method.After general 10 two field pictures, X _Min ⁿ, X _Max ⁿAnd X _b ⁿPromptly tend towards stability.

When frame signal is effective, receive that whenever a pixel promptly judges,

If grey scale pixel value is less than X _Min ^N-1, min_counter ⁿAdd one;

If grey scale pixel value is less than X _Max ^N-1, max_counter ⁿAdd one;

If grey scale pixel value is less than X _b ^N-1, b_counter ⁿAdd one;

After a two field picture has transmitted, min_counter ⁿ, max_counter ⁿ, b_counter ⁿWrite down in the n two field picture gray-scale value less than X _Min ^N-1, X _Max ^N-1, X _b ^N-1Number of pixels.

Finish iterative operation in the mode of state machine then.

Finite state machine (FSM) is a simple mathematical pattern, has the finite aggregate of discrete type input.With the finite state that order the determined set that is accepted input by a basis, finite state machine can have the output of a finite aggregate.If like this, state machine will produce a series of output to reflect a series of input.

Finite state machine (FSM) is defined as five-tuple a: M={I, O, S, δ, λ }, I is that the finite nonempty set of input is closed (input can be a vector); O is that the finite nonempty set of output is closed (output can be vector); S is that the finite nonempty set of state is closed; δ: S * I → S is status change function (State TransitionFunction); λ: S * I → O is output function (Output Fuction).At any time, state machine always is in a certain state, and when input entered, state machine then by the effect of status change function, was transformed in another state.

The five-tuple of a state machine of contrast, the present invention designs a template parameter programming state machine with the way of determining five-tuple, its status transition chart such as Fig. 3, wherein:

TS ₀Be initial state, TS ₁Be statistical picture attitude, TS ₂For calculating step-length coefficient (K _Min ⁿK _Max ⁿK _b ⁿ) attitude, TS ₃For calculating step-length (Step _Min ⁿStep _Max ⁿStep _b ⁿ) attitude, TS ₄For calculating waypoint (X _Min ⁿX _Max ⁿX _b ⁿ) attitude.Original state is TS ₀, this state machine

Be input as: reset signal, frame useful signal, min_counter ⁿ, max_counter ⁿ, b_counter ⁿ

Be output as: X _Min ⁿ, X _Max ⁿ, X _b ⁿThe value of waypoint;

Output function is: the value of calculating and export waypoint;

TS ₀Attitude: after system logic resetted, state was TS ₀, X _Min ⁰, X _Max ⁰Get than the initial value near expectation value, X _b ⁰Get (X _Min ⁰+ X _Max ⁰)/2; Jump to TS after end and the frame signal of resetting is effective ₁Attitude;

TS ₁Attitude: add up in the n two field picture gray-scale value less than X _Min ^N-1, X _Max ^N-1, X _b ^N-1Number of pixels, and be recorded in min_counter respectively ⁿ, max_counter ⁿ, b_counter ⁿIn.When frame signal is invalid, jump to TS ₂Attitude;

TS ₂Attitude: calculate step-length coefficient (K _Min ⁿK _Max ⁿK _b ⁿ), after finishing, calculating jumps to TS ₃Attitude;

TS ₃Attitude: calculate step-length (Step _Min ⁿStep _Max ⁿStep _b ⁿ), after finishing, calculating jumps to TS ₄Attitude;

TS ₄Attitude: iteration waypoint (X _Min ⁿX _Max ⁿX _b ⁿ), after finishing, calculating jumps to TS ₅Attitude;

TS ₅Attitude: Idle state, output waypoint (X _Min ⁿX _Max ⁿX _b ⁿ) value, when frame signal is effective, jump to TS ₁Attitude;

So, the waypoint obtaining step will be under the control of internal state machine stable carrying out.When frame signal is effective at TS ₁Attitude statistical picture characteristic; When frame signal is invalid: at TS ₂Attitude, TS ₃Attitude, TS ₄Attitude iteration waypoint is finished iteration.At TS ₅Attitude output waypoint (X _Min ⁿX _Max ⁿX _b ⁿ) value and wait for the next round iteration.

(3) segmentation stretches

Two sections drawing process are established wherein X _In ^N+1Be the original gray scale of n+1 frame input picture, X _Out ^N+1Be the gray scale that segmentation stretches and exports behind the n+1 two field picture, S1 is for distributing to [X _Min ⁿ, X _b ⁿ] gray level, S2 is for distributing to [X _b ⁿ, X _Max ⁿ] gray level.Make S1=200, S2=55.16 original image is quantized to 8.

The X that n frame iteration obtains before utilizing _Min ⁿ, X _Max ⁿAnd X _b ⁿThe n+1 two field picture stretches.Transformation for mula is:

$X_{\mathrm{out}}^{n+1}=\left\{\begin{array}{cc}0& \&ForAll;X_{\mathrm{in}}^{n+1}<X_{\min }^n\\ S1*(X_{\mathrm{in}}^{n+1}-X_{\min }^n)/(X_b^n-X_{\min }^n)& \&ForAll;X_{\min }^n<X_{\mathrm{in}}^{n+1}<X_b^n\\ S2*(X_{\mathrm{in}}^{n+1}-X_b^n)/(X_{\max }^n-X_b^n)& \&ForAll;X_b^n<X_{\mathrm{in}}^{n+1}<X_{\max }^n\\ S1+S2& \&ForAll;X_{\mathrm{in}}^{n+1}>X_{\max }^n\end{array}\right.$

The combination property of table 1 expression histogram equalization method, one section linear stretch method, the inventive method relatively.If the image size is 320*256, the pixel bit wide is 16bits;

The comprehensive comparison of three kinds of algorithms of table 1

Compare index	The histogram equalization method	One section linear stretch method	The inventive method
				Expend storage resources	65536*17bits＝ 1,114,112bits	Do not have	6*320*16bits＝ 30720bits
The stretching dirigibility	Fine	Very low	Better

The filtering of salt-pepper noise

Do not have

Do not have

Better

Take all factors into consideration three kinds of stretching algorithms of in FPGA, realizing, though histogram equalization can reach good drawing effect, but the storage resources that expends is too high, one section linear stretch has solved storage resources well and has expended excessive problem, but because its iterative algorithm efficient is lower, can only obtain the maximal value and the minimum value of image, carry out one section linear stretch.The inventive method has been inherited the characteristics of one section low storage resource consumption of linear stretch, carries out segmentation and stretches, and develops the interframe iterative algorithm based on negative feedback thought, stably obtains every section waypoint rapidly, has reached better image and has strengthened effect.

Claims (2)

1. method for strengthening infrared focal plane array image, its step comprises:

The 1st step, medium filtering:

The n two field picture is carried out medium filtering, and n is the frame number of pending image;

In the 2nd step, waypoint obtains:

Utilize following formula (I), (II) and formula (III) to calculate the minimum gradation value of the n two field picture of process medium filtering respectively Maximum gradation value

Gray-scale value with waypoint

$\left\{\begin{array}{c}{X}_{\min }^n=X_{\min }^{n-1}+{\mathrm{Step}}_{\min }^n\\ {\mathrm{Step}}_{\min }^n=K_{\min }^n*\mathrm{\&Delta;X}\\ K_{\min }^n=M*N*\min \%-\min \_\mathrm{counte}{r}^n\end{array}\right.$ Formula (I)

$\left\{\begin{array}{c}{X}_{\max }^n=X_{\max }^{n-1}+{\mathrm{Step}}_{\max }^n\\ {\mathrm{Step}}_{\max }^n=K_{\max }^n*\mathrm{\&Delta;X}\\ K_{\max }^n=M*N*\max \%-\max \_\mathrm{counte}{r}^n\end{array}\right.$ Formula (II)

$\left\{\begin{array}{c}{X}_b^n=X_b^{n-1}+{\mathrm{Step}}_b^n\\ {\mathrm{Step}}_b^n=K_b^n*\mathrm{\&Delta;X}\\ K_b^n=M*N*b\%-b\_\mathrm{counte}{r}^n\end{array}\right.$ Formula (III)

With

Be respectively the gray-scale value of minimum gradation value, maximum gradation value and the waypoint of n-1 two field picture, M*N is the resolution of n two field picture, min%, max% and b% represent respectively in the n two field picture gray-scale value less than

With Number of pixels account for the ratio of total number of pixels, wherein, the value of min and max and b is preestablished by the user; Min_counter ⁿ, max_counter ⁿAnd b_counter ⁿBe respectively in the n two field picture gray-scale value less than

With

Number of pixels, Δ X is the iteration weights;

In the 3rd step, segmentation stretches:

Utilize the 2nd step iteration to obtain

With

The gray-scale value that the n+1 two field picture that stretches, segmentation stretch and export behind the n+1 two field picture

$X_{\mathrm{out}}^{n+1}=\left\{\begin{array}{cc}0& \&ForAll;X_{\mathrm{in}}^{n+1}<X_{\min }^n\\ S1*(X_{\mathrm{in}}^{n+1}-X_{\min }^n)/(X_b^n-X_{\min }^n)& \&ForAll;X_{\min }^n<X_{\mathrm{in}}^{n+1}<X_b^n\\ S2*(X_{\mathrm{in}}^{n+1}-X_b^n)/(X_{\max }^n-X_b^n)& \&ForAll;X_b^n<X_{\mathrm{in}}^{n+1}<X_{\max }^n\\ S1+S2& \&ForAll;X_{\mathrm{in}}^{n+1}>X_{\max }^n\end{array}\right.$

Wherein

Be the original gray scale of n+1 frame input picture, S1 is for distributing to

Gray level, S2 is for distributing to

Gray level, S1, S2 is set by the user, the gray level that S1+S2 need be stretched to for the user.

2. method for strengthening infrared focal plane array image according to claim 1 is characterized in that: the 1st step selected for use the 3*3 square window to carry out medium filtering, and at first three column elements to a 3*3 window sort; Three row elements for the window after the row ordering sort then; Ask the intermediate value of three elements on the diagonal line at last, the gained intermediate value is exactly the intermediate value of 9 elements.

Images