CN103283215B - Automatically focusedimage system - Google Patents

Abstract

The present invention discloses a kind of automatic focusedimage system, and it includes a pel array being coupled to a focus signal generator.This pel array catches an image with multiple edges.This generator produces a focus signal, and this focus signal is measured with multiple edge sharpness and become, and each edge sharpness tolerance is to record from one of multiple edge such as this.This generator can not have sufficient reflection symmetry based on an edge is detected, to determine the relative extent reducing the contribution to this focus signal for this edge according to a predefined criterion in a gradient sequence of the picture signal on this edge.

Description

Automatically focusedimage system

Inventor

Hiok Nam TAY

The cross reference of related application

Subject application advocates the excellent of PCT/IB2010/055641 PCT Patent Application case filed in 7 days December in 2010 First weigh.

Technical field

Disclosed subject matter relates generally to automatically focus on the image of electron capture.

Background technology

The photographic equipments such as such as digital camera and digital camera can comprise electronic image sensor, its catch light for It is processed into static or video image respectively.Electronic image sensor generally comprises millions of light capturing elements, such as photoelectricity two Pole pipe.

Many image capture devices such as such as camera include autofocus system.Automatically focusing comprises the following steps： Catch image；Process image to determine if to focus；And if do not focused, then produce in order to change the position of condenser lenses Put the feedback signal of (" focal position ").There are two kinds of main Techniques of Automatic Focusings.The first technology is related to contrast measurement, separately One technology considers the phase contrast between a pair of image.In contrast method, analyze the intensity difference between neighborhood pixels, and adjust Focus, till detecting maximum-contrast.Although being acceptable for static images, contrast technology for It is inappropriate for sport video.

Phase difference method includes for incoming image splitting into two images, and it is by single image capture sensor.By two Individual image is compared to determine phase contrast.Adjustment focal position, till two images match.Phase difference method needs volume Outer part, such as beam splitter and additional image sensor.In addition, the relatively small fixed test point of phase difference method analysis Band.There is less test point group be easy to lead to error because noise stackable on one or more aspects.As detection Point is not overlapped with image border, then this technology is also invalid.Finally, because phase difference method, by photo-fission, is therefore clashed into The amount of the light on optical sensor is cut to two halves or more.In the relatively low dim ambient of image light intensity, this can become For problem.

Content of the invention

The present invention discloses a kind of automatic focusedimage system, and it includes being coupled to the pel array of focus signal generator. This pel array catches image, and this image has multiple edges.This generator produces focus signal, and this focus signal retinue should The multiple edge sharpness recording etc. each of multiple edges are measured and are become.This generator can be according to a predefined criterion, base In an edge is detected, there is no in a gradient sequence of the picture signal on this edge sufficient reflection symmetry, come really Surely reduce a relative extent of the contribution to this focus signal for this edge.

Brief description

Fig. 1 is the schematic diagram of the embodiment of automatic focusedimage pick device；

Fig. 2 is the schematic diagram of the alternate embodiment of automatic focusedimage pick device；

Fig. 3 is the block diagram of focus signal generator；

Fig. 4 is the explanation of the computing of horizontal Sobel Operator on picture signal matrix；

Fig. 5 explanation calculates border width from horizontal gradient；

Fig. 6 A, Fig. 6 B are the explanations of the calculating of the border width of the vertical edge with angle of inclination φ；

Fig. 6 C, Fig. 6 D are the explanations of the calculating of the border width of the horizontal edge with angle of inclination φ；

Fig. 7 is the border width calculating angle of inclination φ and correcting the vertical edge with gradient；

Fig. 8 is the explanation vertically cascading edge；

Fig. 9 A is the explanation of the group of closelypacked vertical bar；

Fig. 9 B is the curve chart of the picture signal crossing Fig. 9 A；

Fig. 9 C is the curve chart of the horizontal Sobel gradient crossing Fig. 9 A；

Figure 10 is the flow chart eliminating the process at tightly packed edge with shallow modulation depth；

Figure 11 is explanation for calculating the rectangular histogram of the border width of border width scope of fine focusing signal；

Figure 12 is the explanation of scene；

Figure 13 is the curve chart of the change of narrow edge counting focusing on during scanning of the scene that Figure 12 is described；

Figure 14 is the curve chart of the change of hair focus signal focusing on during scanning of the scene that Figure 12 is described；

Figure 15 is the curve chart that the change of fine focusing signal in focal position scope is described；

Figure 16 be show multiple objects in scene and this etc. the saying of the equipment of selected marker on one of object Bright；

Figure 17 is the block diagram of the alternate embodiment of focus signal generator；

Figure 18 is the schematic diagram of the alternate embodiment of automatic focusedimage pick device；

Figure 19 is the embodiment of the automatic focusedimage pick device with main pixel array and auxiliary pixel array Schematic diagram；

Figure 20 is that the replacement of the automatic focusedimage pick device with main pixel array and auxiliary pixel array is implemented The schematic diagram of example；

Figure 21 is that the replacement of the automatic focusedimage pick device with main pixel array and auxiliary pixel array is implemented The schematic diagram of example；

Figure 22 is from the change of the border width of main pixel array and from auxiliary picture in different focal position The explanation of the change of the border width of pixel array；

Figure 23 A explanation picture signal is on two neighboring edges of the noninteracting relative polarity in edge (i.e. sign) Gradient；

Figure 23 B explanation picture signal is on two neighboring edges of the relative polarity (i.e. sign) that edge interacts Gradient；

Figure 24 A shows the positive gradient from Figure 23 B, and illustrates between a pair interpolated gradient at specific graded levels Distance is divided into two non-equally by interpolated peak；

Figure 24 B show is derived from the positive gradient of Figure 23 B, and illustrates between two graded levels and in left side and right side by gradient The region that the both sides of profile are delimited is divided into the Liang Ge area of unequal-area by the vertical line immediately below interpolated peak；

Figure 24 C shows the positive gradient from Figure 23 B, and the length of the fragment of gradient profile between two graded levels is described The area in the area below degree, this fragment and lower gradient horizontal vertical, and the width of the substrate in this area；

Figure 24 D shows the positive gradient from Figure 23 B, and the method for estimating the first derivative is described；

Figure 24 E shows the positive gradient from Figure 23 B, and the alternative method for estimating the first derivative is described；

Figure 24 F show is derived from the positive gradient of Figure 23 B, and the method for estimating flection is described；

Figure 24 G shows the positive gradient from Figure 23 B, and the alternative method for estimating flection is described；

Figure 24 H show is derived from the positive gradient of Figure 23 B, and illustrates between the midpoint at different graded levels and each midpoint The distance between with interpolated peak；

Figure 24 I shows the positive gradient from Figure 23 B, and two gradients away from interpolated peak mutual distance are described；

Figure 24 J is shown symmetric gradient profile and so that midpoint is overlapped with interpolated peak；

The distance of multiple of spacing between comparison is as continuous flection for Figure 25 explanation and on the edge drawn The flection sequence of picture signal, it shows the width W between (a) a pair of posivtive spike and negative peak_s, (b) has given value h₁'s Width W between the interpolated flection of a pair of outermost₁, (c) has given value h₁Interpolated within a pair Width W between two derivatives₂, and (d) from (this is between posivtive spike and negative peak) zero passage to having given value h₁Outermost Interpolated flection apart from D₁；

The figure of the picture signal that Figure 26 explanation is drawn in comparison with the distance of the multiple as the spacing between continuous sample As data sample sequence, it shows the width W between two samples at the two ends at (a) edge_edgeAnd contrast C_edge, (b) has Peak Grad g between a pair of sample that the steepest having sample value changes_peak, (c) edge there is contrast C₁And width W_part1Non- branch divide, and (d) edge have contrast C₂With width W_part2Non- branch divide；

The gradient on edge that Figure 27 explanation is drawn in comparison with the distance of the multiple as the spacing between continuous sample The area in the area under sequence and drawn gradient sequence；

Figure on the edge that the distance of the multiple in comparison with the spacing between as continuous gradient is drawn for Figure 28 explanation Gradient sequence, a center of gravity (that is, the centre of moment) and this constant gradient distance away from this center of gravity as signal；

The position at interpolation peak is found in Figure 29 explanation by interpolation；

Figure 30 shows the alternate embodiment of focus signal generator.

Specific embodiment

Disclosed is a kind of automatic focusedimage system, and it includes being coupled to the pel array of focus signal generator. This pel array catches image, and this image has at least one edge of a width.Focus signal generator can produce focusing Signal, its statistics with border width and/or border width and become.Disclosed is a kind of automatic focusedimage system, its bag Include the pel array being coupled to focus signal generator.This pel array catches image, and this image has a width at least One edge.This generator produces focus signal, and this focus signal becomes with the various statistics of border width and border width.Should Generator can eliminate the edge of the unsymmetry of the gradient with picture signal.This generator also can eliminate due to the phase in gradient Correlation peaks and make the edge of die plate failure.Processor collectiong focusing signal and/or the statistics of border width, and adjust focusing thoroughly The focal position of mirror.Border width can be determined by include using the various technology including gradient.The straight of border width can be used Square figure come to determine specific image be focus on or do not focus on.The rectangular histogram instruction with big thinner border width counting focuses on Image.

Framework

Come referring more specifically to schema by reference to numeral, Fig. 1 shows that automatic focusedimage catches the enforcement of system 102 Example.System 102 can be a part for Digital Still Camera, it will be understood that, this system can appointing in the controlled focal needing image What implement in device.System 102 may include condenser lenses 104, pel array and circuit 108, A/D converter 110, processor 112nd, display 114, memory card 116 and drive motor/circuit 118.Light from scene enters through lens 104. Pel array and circuit 108 produce analogue signal, and analogue signal is converted to digital signal by A/D converter 110.Pel array 108 may be incorporated into and inlayed colour image, such as Bayer pattern.Digital signal can be sent to processor 112, it executes various places Reason, the such as control of color interpolation, focal position, colour correction, image compression/decompression, user interface control and display Control；And digital signal can be sent to focus signal generator 120.In focus signal generator 120 and processor 112 resident In the case of in different encapsulation, color interpolation unit 148 can be implemented so that color interpolation is executed to digital signal 130, think poly- Burnt signal generator 120 estimates the missing color signal in each pixel.Or, in focus signal generator 120 and processor In the case that 112 reside in encapsulation 144 together, focus signal generator 120 can be as shown in Fig. 2 input in bus 146 From the interpolated color image of processor 112, or from the original image signal being produced by A/D converter 110 draw single Picture signal, such as gray-scale signal.

Focus signal generator 120 receives one group of control signal 132 from processor 112, in addition, and can be by signal 134 export processor 112.Output signal 134 may include one of the following or one or more of：Focus signal 134, narrow The set of number of the statistics of border width in edge counting and representative image.Processor 112 can produce focus control signal 136, it is sent to drive motor/circuit 118 to control condenser lenses 104.The image of focusing can finally be provided to show Device 114 and/or be stored in memory card 116.Algorithm for adjusting focal position can be executed by processor 112.

Pel array and circuit 108, A/D converter 110, focus signal generator 120 and processor 112 can all be stayed Exist in an encapsulation.Or, pel array and circuit 108, A/D converter 110, focus signal generator 120 can be independent of places Manage device 112 and reside in encapsulation 142 as the imageing sensor 150 shown in Fig. 1.Or, focus signal generator 120 He Processor 112 can reside in encapsulation 144 as camera controller 160 independent of pel array 108 and A/D converter 110. Focus signal generator 120 (or any alternate embodiment, such as the embodiment shown in Figure 30) and processor 112 can together with stay Exist in semi-conductive substrate (such as silicon substrate).

Focus signal generator

Fig. 3 shows the embodiment receiving the focus signal generator 120 of image from image providing unit 202.Image There is provided unit 202 can be the color interpolation device 148 in Fig. 1 or the processor 212 in Fig. 2.Focus signal generator 120 may include Rim detection and width measure (EDWM) unit 206, focus signal computer 210, lengthy filter 212 and width filtering Device 209.Focus signal generator 120 can further include to refine switch 220, and it is controlled by inputting " refinement " 222.Focus signal Generator 120 can provide the narrow edge from spanning filter 209 to count and the focusing letter from focus signal computer 210 Number, this focus signal can configure between fine focusing signal and hair focus signal, and it can be selected by input " refinement " 222.Or, Fine focusing signal and hair focus signal, and the part output as output signal 134 can be calculated.Rim detection with Width measure unit 206 receives the image being provided by image providing unit 202.In the context of Fig. 1 and 2, control signal (example As control signal " refinement " 222) can be provided in signal 132 by processor 112.And in the situation of Fig. 1 and Fig. 2, can be by Output signal 134 is supplied to processor 112, and processor 112 serves as focusing system controller, and it controls the poly- of condenser lenses 104 Burnt position, to make the image of object reach pel array by analyzing output signal 134 with the sharp object in detection image Sharp focusing on 108.The various assemblies of focus signal generator 120 are described below.

The convertible input picture of EDWM unit 206 is so that three signals red (R) of image, green (G) and indigo plant (B) are changed For single image signal.Using several technology, image can be transformed to single image.Can be calculated using rgb value brightness or Chromatic value, or the specific ratios of desirable rgb value are forming single image signal.For example, available equation Y=0.2126*R+ Calculating brightness value, wherein Y is brightness value to 0.7152*G+0.0722*B.This single image signal can be then by Gaussian filter Or any low pass filter is processed, noise is removed with the view data sample value between smooth neighbor.

Focus signal generator 120,120', 120 " be not limited to gray-scale signal.Focus signal generator 120,120', 120 " can be to any one image signal manipulations, to detect one of this picture signal or more than one edge.Or, focus on Signal generator 120,120', 120 " any combinations of picture signal can be operated, such as Y, R-G or B-G.Focus signal produces Device 120,120', 120 " can respectively to each of R, G, B picture signal or its any one or more than one combination operation, To detect edge.Focus signal generator 120,120', 120 " each of R, G, B picture signal or its any group can be directed to Close and form the statistics of border width.Focus signal generator 120,120', 120 " can be according to from one or more figures Statistics as the border width of signal forms focus signal.

Focus signal generator comprises edge detector to identify the edge in picture signal.Edge detector can use one Rank edge detection operator, such as Sobel (Sobel) operator, Puri Wei Te (Prewitt) operator, Robert (Roberts) hand over Fork operator or Robert's operator.Edge detector can identify edge using the edge detection operator of higher-order, and such as second order is calculated Son, such as Laplace operator.Edge detector can use any one of known edge detection operator or also use known operator Any one of usual rim detection principle any improved operator

In the case that edge detector uses single order edge detection operator, the gradient calculating picture signal (first is led Number).Various methods can be used for calculating gradient, comprises using any one of various single order edge detection operators, such as Sobel Operator, the special operator of Puri dimension, Robets cross operator and Robert's operator.Robert's operator has two kernels, and it is single-row Or row vector：[- 1+1] and its transposition.Robets cross operator has two kernels, and it takes advantage of 2 matrixes for 2：[+1,0；0,-1] [0 ,+1；- 1,0], with [the first row vector；Second row vector；The third line vector] form show, such as one in Matlab Sample.The special operator of Puri dimension and Sobel Operator substantially have same kernel [- 1,0 ,+1], and it takes the gradient on line direction, and Its transposition takes the gradient on column direction, is multiplied by the different low pass filtered executing the low-pass filtering perpendicular to corresponding gradient direction further Ripple device kernel.The gradient in columns and rows for example can be calculated respectively using Sobel X operator and Sobel Y operator, to examine respectively Survey vertically and horizontally edge.Provide the Suo Bei at location of pixels [k, q] place by equation Sx [k, q]=U [k, q+1] U [k, q-1] That X operator, wherein k is line number, and q is column number.Be given same by equation Sy [k, q]=U [k+1, q] U [k-1, q] Sobel Y operator at position, wherein U is the picture signal of treated image.

In the case that edge detector uses Second-order Operator, calculate flection (the such as La Pula of picture signal This).

Orientation tags

If horizontal or vertical gradient magnitude exceedes predetermined lower bound (" elimination threshold value ") (being for example directed to 8 bit images is 5), that Each pixel can be added with the label of horizontal edge (" H ") or vertical edge (" V "), or if being neither, then plus no The label at edge.The pseudo-edge that the elimination of this lower limit leads to because of soft shadows or noise.If the horizontal gradient value of pixel surpasses Cross predetermined hysteresis of its perpendicular gradient magnitudes or more (being for example, 2 for 8 bit images), then can be by pixel mark-on Sign as vertical edge, and vice versa.If two gradient magnitude differences are less than hysteresis, then pixel obtains nearest with it There is the direction label identical direction label of the neighbor of fixed direction label.For example, if in each row Scanogram line by line from left to right and downwards, then the inspection sequence of neighbor can be topmost pixel first, upper left side picture Element second, and left pixel the 3rd, and upper right side pixel is last.This is applied delayed to contribute to guaranteeing if every in neighborhood pixels One has almost identical horizontal and vertical gradient magnitude, then neighborhood pixels obtain similar label.Fig. 4 illustrates level and hangs down Tagged result on the 6x6 array of vertical ladder degree.In each unit, horizontal gradient be located at upper left, vertical gradient on right side, And direction label is in bottom.In this step, the pixel only with horizontally or vertically gradient magnitude more than 5 is qualified, because side Edge pixel is with bold print, and obtains direction label.

Can be for vertical edge horizontal sweep and for horizontal edge vertical scanning image, gradient and label.If same Have in row identical horizontal gradient polarity and all for vertical edge the left side of tagged each contiguous pixels group or The neighborhood pixels on right side are not same, then this pixel group can be denoted as vertical edge.Equally, if having phase in same row With vertical gradient polarity and all neighbouring above or below tagged each contiguous pixels group for horizontal edge Pixel is all unsatisfactory for above-mentioned condition, then this pixel group can be denoted as horizontal edge.Therefore, can recognize that horizontal and vertical side Edge.

Border width

Can by remove gradient magnitude be less than intramarginal peak gradient magnitude given fraction pixel become more meticulous each Edge.Fig. 5 illustrate using the peak gradient magnitude equal to edge 1/3rd become more meticulous threshold value come by border width from original 9 become more meticulous downwards 3 this step.This edge becomes more meticulous and distinguishes that (it is set in the acutance at edge to dominant gradient component Visually-perceptible in dominant apparent border width), but regardless of tool is likely to result in gradient and mildly declines in many pixels The image of the multiple overlap shades moving back.

Border width can be calculated in any one of known method.A kind of method calculating border width is simply opposite side The number of the pixel in edge is counted.The alternative method calculating border width is shown in Fig. 5.In Figure 5, by from becoming more meticulous Threshold value 304 carries out interpolation, looks between the first exterior pixel (pixel 3) through refinement edge and neighbouring outside pixel (pixel 2) To the first fractional pixel position (2.4).Equally, between the second exterior pixel (pixel 5) outside adjacent thereto pixel (pixel 6) Find the second fractional pixel position (5.5).Think that border width is the difference between this two fractional pixel position, 5.5-2.4= 3.1.

Another replacement border width computational methods are to calculate picture signal edge (becoming more meticulous with or without edge) On difference, and by its divided by edge peak gradient.

Or, border width can be (or interpolated for a pair of posivtive spike of the second dervative of picture signal on edge and negative peak The distance between peak).Other replacement schemes are possible, will be under further in this manual title " edge sharpness tolerance " Description.

Will be seen that there are the other replacements different from width further under title " edge sharpness tolerance " in the present note Scheme, and width is only an example of substantially unrelated with the illumination of scene edge sharpness tolerance.

Gradient corrects

Although each edge can be assigned to an assigned direction (for example vertically or horizontally), or another orthogonal Assigned direction (for example, horizontal direction or vertical direction), and can have in the side orthogonal with the edge direction that is assigned The border width recording upwards, the border therefrom producing between the area of the different images signal value in the image at these edges can not Or it is not generally fully aligned with arbitrary assigned direction.In fig. 6, by border (shadow band) be shown as with respect to vertical dotted line with Inclination angle phi tilts, and width a is shown as recording on orthogonal direction (that is, horizontal direction).However, in the direction with border The width b (as indicated in the figure) recording on (being also the direction at the edge of a part forming border) orthogonal direction compares width A is more suitable for the width at border (and edge).It is not this little width a often mistakes of recording orthogonal with corresponding edge direction Greatly, and not represent the actual thickness on corresponding border.

For the purpose calculating focus signal from border width, one of those assigned directions or another one record Border width will be corrected by the width being reduced on the direction orthogonal with the direction of respective edges.Edge Detection executes this correction to border width with width measure unit 206.As shown in FIG, the width a recording is right angle three The length of angular hypotenuse, the base (with width b labelling) of this right angled triangle orthogonally across on shadow edge (therefore with side Edge direction is orthogonal), and this right angled triangle has angle φ.Then from the width a recording to the direction orthogonal with edge direction Projection obtains calibrated width b.Knowable to base triangle, this projection can be given by b=a cos (φ), but as long as approximately Accuracy within 20% for the method acquisition is so that it may use method of approximation.Angle φ or cos (φ) itself can be by known in technique For finding out any method in the direction at the edge in image, or by more accurate described in the flow chart shown in Fig. 7 Method finding out.

The border width at each horizontally or vertically edge can be (specified from being horizontally or vertically oriented respectively for its gradient Direction) it is corrected.Fig. 6 A, 6B explanation is for the border (and therefore forming the edge on this border) with vertical line with gradient The correction of the border width recording in the horizontal direction calculates.Fig. 6 C, 6D explanation is for the border with horizontal line with gradient The correction of the border width that (and therefore forming the edge on this border) records in vertical direction calculates.Can be by making in designated parties The border width recording on (for example vertically or horizontally) is multiplied by factor cos φ to carry out this correction, and wherein φ is Angle of inclination with assigned direction.

For example, Fig. 7 shows the mistake of calibration edge width in order to the gradient for the edge tilting from vertical line The flow chart of journey.(for horizontal edge, in flow charts, replaced " arranging " with " OK ", and " vertical " is exchanged with " level ".)

From step 502 to step 506, find out angle of inclination φ.For each vertical edge, at step 502, positioning is hung down Vertical ladder metric reaches the column position of peak value, and finds out horizontal gradient x.At step 504, find out perpendicular gradient magnitudes along row position Put and reach peak value in two separate pixels, and find out vertical gradient y.

At step 506, find out angle of inclination φ=tan^-1(y/x).At step 506, can by search look-up table Find out angle of inclination.Although step 502 to 506 assumes a kind of specific program finding out angle of inclination and method, can be changed to make With other programs known in the art and method.

Finally, at step 508, be multiplied scaled border width by with cos (φ) or with its approximate representation, As those skilled in the art generally does in practice.

First modification of the process shown in Fig. 7 is looking into of the entry by providing the various combinations with input value x and y Table is looked for replace the part of step 506 and step 508.For each combination of input value x and y, look-up table returned edge width Degree correction factor.The border width correction factor being exported by look-up table can be cos (tan^-1(y/x)) within 20% (preferably Within 5%) approximate representation.Then border width is made to be multiplied by this correction factor, to produce the edge width through gradient correction Degree.

Second is revised as calculating the business y/x between vertical gradient y and horizontal gradient x, to produce business q, is then used by q next defeated Enter the look-up table of the entry to the various values having for q.For each value of q, the look-up table returned edge width calibration factor. Border width correction factor can be cos (tan^-1(q)) within 20% (preferably within 5%) approximate representation.

In order to find out angle of inclination φ (or its approximate representation is so that correction factor is accurate within 20%) and subsequently look for Go out correction factor cos (φ) (or its approximate representation), or do not find out angle of inclination φ in order to directly find out correction factor (such as to exist In first and second modifications), the value of x and y can be obtained in step 502 to 506, but can be changed to using other methods.

3rd modification is operates following to the execution of each of multiple pixels in edge：A () finds out level for pixel Gradient x and vertical gradient y, (b) finds out q=y/x for this pixel, and (c) finds out the correction factor corresponding to q, for example cos(tan^-1(q)) or its approximate representation within 20%.Finally, by the correction from each of multiple pixels The factor is averaging the correction factor to find out for this border width.Meansigma methodss can be weight averaged value, for example, wherein have The pixel of greater level gradient is given than has the meansigma methodss of the big weight of another pixel of less horizontal gradient.

Other modifications along these directions or other direction are possible.

Screening threshold value

If the peak gradient magnitude of neighboring edge is less than the predetermined score of the neighbouring peak gradient magnitude of relatively broad edge, then Can stop this grade neighboring edge that focus signal is contributed completely, or make it contribute weakening.Fig. 9 A, Fig. 9 B and Fig. 9 C explanation normal solution Problem certainly.

Two narrow black three vertical white bands being spatially separating that Fig. 9 A explanation is 2 pixels by respective width.Middle informal voucher Width be 2 pixels of fillet.Fig. 9 B show level cross the image in Fig. 9 A in sharp image and broad image Each and describe picture signal.The Sobel x gradient for sharp image and broad image of Fig. 9 C depiction 9B.In figure In 9C, as expected, the first edge (pixel 2 to 5) of broad image is wider than sharp image, and final edge (pixel 13 to 15) It is also such.However, two narrow edges (pixel 9 and 10, and pixel 11 and 12) have width two in both images.? In Fig. 9 B, the corresponding slope at pixel 9 and 10 and pixel 11 and 12 respectively asks for two pixels to complete transition.However, it is fuzzy Image has from compared with broad edge being remarkably decreased to the peak gradient magnitude of narrow side edge, and up to 50%.On the other hand, sharp image It is less than 10% relatively changing between broad edge and narrow edge.

Be adjacent to the narrow side edge of the relatively broad edge with relative sign gradient peak gradient magnitude be remarkably decreased (example As 20% or bigger) the fuzzy image of prompting is without well focussed, and therefore, should not rely on narrow side edge as broad image is Sharp instruction.

Equally, as long as the edge being mutually adjacently with alternating gradient polarity is very close to each other, for example, it is separated by less than 1 Individual pixel (" minimum edge gap "), just should not rely on it to indicate for this, even if its border width is less.Between minimum edge Gap is the number according to pixel, such as 12 or between.

Furthermore, it is assumed that an edge may be eliminated because having less than the peak gradient eliminating threshold value, there is identical ladder Degree polarity and separate less than minimum edge gap twice add sharp_edge_width (sharp_edge_width be through Assign to indicate the numeral of the border width of sharp edges) two continuous boundaries can be used as eliminate or degrade be derived from two One of mutually adjacent edge or the condition of both contributions.Arbitrary.

Rim detection and width measure unit 206 can perform following algorithm, for based on from the sieve set up compared with broad edge Select threshold value and can open and close modulation screening flag and eliminate closelypacked narrow side edge.

For each edge, the process of the flow chart according to Figure 10 come determine by be used for relative polarity immediately under The screening threshold value at one edge and screening flag.

In the case of given screening threshold value and screening flag, edge can be cancelled, unless one of following condition is true： A (), for this edge, is screened flag and is closed, the peak gradient magnitude at (b) edge is not less than the screening threshold value for this edge.Can be by Condition (c) is added to condition (a) and (b), and border width is not less than sharp_edge_width+1, has been wherein sharp_ Edge_width assigns a numeral to indicate the border width of sharp edges, and wherein can change "+1 " to be set higher than sharp_ The scope of the border width of edge_width, within the range, if edge is unable to reach (a) and (b), then eliminate this equilateral Edge.For the example shown in Fig. 9 A to Fig. 9 C, sharp_edge_width can be 2.

Figure 10 is the flow chart determining screening threshold value and screening flag for each edge.For vertical edge it is assumed that going in edge From left to right scan, but what this was not required.(for horizontal edge it is assumed that scanning from top to bottom along row, but this is not required 's.) assign a numeral for sharp_edge_width, and it can be 2 for the example shown in Fig. 9 A to Fig. 9 C.In step 702 Place, starts at first edge, at step 720, inquires whether its border width of each edge adds more than or equal to one Sharp_edge_width, value one is the minimum edge gap width illustrating for this, but can be using different values, such as 0.5 And between 2.0.If it is then this edge is compared with broad edge, and step 706 then will have the immediately lower of relative polarity The screening threshold value of edge is set as that β is multiplied by the peak gradient magnitude at edge, and β is the then step 708 from 0.3 to 0.7, preferably 0.55 Open the screening flag for next edge, proceed to next edge.If not, so this edge is not compared with broadside Edge, and step 730 then checks that the spacing at the previous edge away from identical Polymer brush is whether (or different more than minimum edge gap Predetermined number) twice add sharp_edge_width, and the immediate previous edge (if present) of relative polarity Whether more than away from minimum edge gap.If it is then step 710 is then switched off the screening flag for next edge. If not, so keeping the screening flag for next edge and screening threshold value, and proceed to next edge.β can be predetermined point Count, or it can be the fraction calculating according to predetermined formula, the function of such as border width.In the latter cases, β can be from image A part arrives another part.

Alternate embodiment

The orientation of pixel grid：

The image being focused signal generator 120 input can have the square of 45 degree of rectangular frame rotation being located relative to image Pixel in shape grid (" pixel grid ").In the case, the X-direction of edge detecting operation and width measure operation and Y side To can similarly rotate.

Edge sharpness is measured：

In the above description, the acutance of the image at edge be by from cross over this edge gradient sequence measured by get Come to represent, wherein this constant gradient is oriented to cross over this edge the width at edge.There is the alternative relying on similar principles work Case.Substantially it is allowed to the reason produce focus signal in this way is, individual rim contribute to and makes view data in proportion Scaling (such as) 20% unrelated or substantially unrelated (change for example under view data scaled 20% and be less than 5%) Amount (hereinafter referred to as " edge sharpness tolerance "), thus compared with conventional comparative's degree detection method, contribute to making focus signal with The illumination of the scene of image or scene in object reflectance unrelated or very little is related.

In the focus signal generator 120 of the present invention, have scaled with the 20% of view data unrelated or real In matter, any edge sharpness tolerance of unrelated above-mentioned characteristic is also another to same Grad from a gradient or interpolated gradient The good replacement of the width that one gradient or interpolated gradient record.

The edge sharpness tolerance substituting preferably has the unit not comprising energy unit.Determine side based on following 2 points The unit of edge sharpness metric：A each sample of the view data to its computing for () single order edge detection operator has energy unit, B the distance between () sample has long measure.Based on point (a) and (b), the unit of Grad is energy unit divided by length list Position.Similarly, the contrast that the arbitrary non-branch on edge or edge divides has energy unit.Therefore, contrast is not Edge sharpness tolerance because this unit manifests it being affected by the illumination of scene and the reflectance of object.The peak gradient at edge The edge sharpness tolerance being not, because having energy unit in the unit of peak gradient, thus also indicate that it is shone by scene The change of degree is affected.On the other hand, the peak gradient at edge is good edge sharpness tolerance divided by the contrast at edge, because Its unit is the inverse of long measure.As another example, Grad exceedes the meter of the gradient of a certain predetermined score of peak gradient Number is good edge sharpness tolerance, because this counting is only the distance degree to quantify with the spacing size between continuous gradient Amount, therefore has long measure.

It may be noted here that during producing edge sharpness tolerance, can be from the single order rim detection in order to detect edge Operator is producing gradient, or can produce gradient from different the first derivative operators (that is, gradient operator).For example although Sobel Operator (or even second order edge detective operators, such as Laplace operator) may be used to detect edge, but its kernel letter It is singly that (it is simply lower the same from the orientation of gradient operator by a sample of view data for [- 1 ,+1] and its transposition Originally deduct) Robert's operator, make gained gradient be located at two samples between center.Can measure independent of producing edge sharpness Or one or more derivative operators used in any one of shape measurements described in next part, with than single order The edge detection operator of high-order is detecting edge.

Viewed another way, the unit of edge sharpness tolerance should be the power of long measure, such as long measure square, long The square root of the inverse, long measure itself or long measure of degree unit.

Arbitrary such edge sharpness that substitutes measures the border width in all alternative focus signal generator 120.

In order to correct to the inclination at edge, should change above with reference to described in Fig. 6 A to 6D and Fig. 7 correction factor (under Literary composition is referred to as " the width calibration factor ") with using same power.For example, if edge sharpness is measured as peak gradient divided by contrast Degree, that is, its unit is the inverse of gage length unit, then the suitable correction factor for edge sharpness tolerance is to join above Examine the inverse of the correction factor described in Fig. 6 A to 6D and Fig. 7.As another example, if the unit of edge sharpness tolerance is long Degree unit square, then for edge sharpness tolerance the slant correction factor should regard the width calibration factor square.

Substitute some examples of edge sharpness tolerance below with reference to the schema description in Figure 27, Figure 28, Figure 25 and Figure 26.

Leap one edge that Figure 27 explanation is drawn in comparison with the distance of the multiple as the spacing between continuous gradient The area A in the shadow region under gradient sequence and the gradient sequence drawn₃.In this example, this area is limited at two ladders Spend horizontal L₁With L₂Between, it can be defined as (example with respect to the interpolated peak Grad (or, peak Grad) of gradient sequence As) predetermined portions of interpolated peak Grad.Shadow region has four corners of interpolated gradient.Area is divided by interpolated peak ladder Angle value (or, peak Grad) it is good edge sharpness tolerance, because it has long measure.Note, the replacement in this area Restriction is possible.For example, this area with gradient sequence rather than can use graded levels L₁Come from upper and delimit.

The image on edge that Figure 28 explanation is drawn in comparison with the distance of the multiple as the spacing between continuous gradient The gradient sequence of the sample of data, center of gravity 3401 (i.e. the centre of moment) and gradient (have Grad g₂、g₃、g₄、g₅And g₆) away from center of gravity Apart from u₂、u₃、u₄、u₅And u₆.Good edge sharpness measures the kth central moment of the gradient around for center of gravity, that is, gradient away from The weight averaged value of the distance of center of gravity, wherein weight are the value of corresponding gradient, and k is even-integral number.For example, k can be 2, It makes edge sharpness measure as variance, and seemingly gradient sequence is that probability distribution is such.In this example, the list of edge sharpness tolerance Position be long measure square.In more general terms, edge sharpness tolerance can be multiple gradients of gradient sequence away from respect to this Multiple gradients and the function of the distance of predefined position, this sequence is the array on this edge.Different from center of gravity, this predefines Position can be the interpolated peak position of gradient sequence.The suitable subset of edge gradient can be selected to participate according to predefined criterion This calculates.For example in order to be sufficiently accurate it may be desired to the Grad of this constant gradient is at least the interpolated peak of this peak gradient or this gradient sequence The predetermined score of Grad.

On the edge that Figure 25 explanation is drawn in comparison with the distance of the multiple as the spacing between continuous flection The flection sequence of the sample sequence of view data, it shows the width W between (a) a pair of posivtive spike and negative peak_s, (b) has Given value h₁The interpolated flection of a pair of outermost between width W₁, (c) has given value h₁A pair within Width W between interpolated flection₂, and (d) from (this is between posivtive spike and negative peak) zero passage to having given value h₁ The interpolated flection of outermost apart from D₁.Can be by three width W of this grade_s、W₁And W₂Any one of be used as edge sharpness Tolerance.

Additionally, in the example of Figure 25, edge sharpness tolerance can be the flection such as this from (this to posivtive spike and negative peak it Between, and can be interpolated) correct one's mistakes zero distance weighted summation, weight therein is the value of corresponding flection.More In general, edge sharpness tolerance can predefine away from respect to multiple flection such as this for the multiple flections on edge The distance of position function.Cross outside zero position, center of gravity is also the good candidate of this predefined position, with weight for this etc. The value of flection.The another good candidate of this predefined position can be for this to the midpoint between positive gradient and negative gradient.

The picture from the edge drawn in comparison with the distance of the multiple as the spacing between contiguous pixels for Figure 26 explanation The sample sequence of the view data of element, it shows the width W between two samples at the two ends at (a) edge_edgeAnd contrast C_edge, (b) has the peak Grad g between a pair of sample of steepest change of sample value_peak(being produced by Robert's operator), (c) Edge there is contrast C₁With width W_part1Zhai Wei branch divide, and (d) edge have contrast C₂And width W_part2Zhai Wei branch divide.As previously mentioned, peak Grad g_peakDivided by contrast C_edgeIt is a good edge sharpness Tolerance.Width W_edgeIt is another good edge sharpness tolerance.Width W_part1And W_part2It is also a good replacement.Can will be right Ratio degree C₁And/or C₂It is defined as contrast on border C_edgePredetermined portions.Or, can by therein any one be defined as edge Peak gradient (for example, peak gradient g_peak) prearranged multiple.It is also noted herein that " Zhai Wei branch divides " can be by the warp of view data Interpolation sample (such as in fig. 26 with square show) or delimited to nearest pixel counts by round down or round-up.

Gradient unsymmetry

Figure 23 A and Figure 23 B illustrates a kind of method, and wherein, focus signal generator detects gradient signal (hereinafter also referred to Gradient profile) in lack peak the symmetry of surrounding, with mitigate or exclude associated edge completely to auto focus control system System (focus signal for example producing via the weight averaged value counting with for example associated border width or edge, but Not limited to this) impact.This peak can be the peak gradient among a series of continuously multiple gradients.Or, this peak can be interpolated Peak gradient, its plural gradient among continuously multiple gradients of this series carrys out interpolation.Figure 23 A explanation is crossed over to be had for a pair The gradient profile of relative polarity (that is, sign) the and adjacent picture signal at edge, this is separate to edge and not phase Interaction.On the other hand, Figure 23 B explanation leap has the gradient wheel of relative polarity the and adjacent picture signal at edge for a pair Exterior feature, this sufficiently closes to edge so that interacting.By comparing 23A and 23B, clearly, when there being the adjacent of relative sign A pair of edge (that is, one of edge such as this has positive gradient, and another one has negative gradient) become close to, it is respective Gradient profile just loses symmetry.

In Figure 23 A, from left to right, gradient profile is increased to posivtive spike gradient 3210 at position 6 in left side, and on the right side Side drops to the negative peak gradient 3260 at position 18.Grad has been standardized to provide peak gradient magnitude 1.0.Respectively each Peak gradient 3210,3260 neighbouring, gradient profile has the left-to-right symmetry with respect to this peak.When using respective peak ladder During 0.3 times of threshold value of metric, positive gradient profile 3211 and negative gradient profile 3261 respectively correspond to the side at respective edge Edge width 5.

On the other hand, in Figure 23 B, in left side in the posivtive spike gradient 3212 at position 6 and negative at position 9 on right side Peak value 3262 ratio is close together in Figure 23 A.In Figure 23 B, such as in Figure 23 A, Grad has been standardized to be given Peak gradient magnitude 1.0.Respectively correspond to positive gradient profile 3213 and two edges of negative gradient profile 3262 show in Figure 23 B So reciprocally effect so that partly offseting each other, thus lead to respectively be located at position 6 with 9 closely adjacent just The reduction of the gradient magnitude between peak 3212 and negative peak 3262.

Therefore, gradient profile 3213,3263 lacks the left and right reflection symmetry on respective edge.Symmetric shortage is in figure In interpolated gradient profile (with solid-line curve show) in especially notable.Can in a certain distance away from this peak gradient (specifically, The distance between 0.2 to 0.7 times of border width between edge) in or its peak graded levels and its non-zero fraction gradient In a certain scope between level (specifically, in 10% to the 90% of peak graded levels；More specifically, arrive 20% In 80%) find symmetric shortage.For example, can be by by the in the distance of the half of border width or between peak ladder The left side of interpolated gradient profile in the range of graded levels between the 20% of degree level and 80% is compared with right side, To find unsymmetry.

As another result, when being used as 0.3 times of peak gradient magnitude of same threshold value to measure, each peak value 3212nd, the border width of 3262 surrounding is reduced to 4.The border width at two edges so recording no longer represents focusing journey Degree.Focus signal generator can detect this unsymmetry, then or cause and mitigate associated border width and give focus signal And/or edge count contribution, even fully phase out it.Say in a broad sense, by the sharpness at the edge in check image Lai The focus control system obtaining more sharp image can mitigate or be completely eliminated from the symmetric edge of gradient profile shortage Impact.

The gradient profile that the isolated multiple edges being produced by the sharp borders in scene in image are had every The one left and right reflection symmetry manifesting respective edge.In Figure 23 A, there is left and right anti-along vertical axis of symmetry (vertical dotted line) Penetrate symmetry, its just below the peak gradient 3210 at position 6 so that along under the reflection of vertical axis of symmetry, at position 4 The gradient at position 7 for the gradient map, and vice versa, and the gradient at position 8 for the gradient map at position 3, and instead As the same, by that analogy.This is because corresponding to the edge corresponding with gradient profile 3211 border just therebetween by Sampling.In many cases, be not the centre on border to the sampling on border, and therefore this vertical axis of symmetry not with location of pixels Overlap, but between two pixels.However, left and right reflection symmetry can be identified using interpolation, as discussed below.

Another Typical gradients profile at the isolated edge of Figure 24 J explanation.Peak gradient 3212' at position 6 and right at it Side is in four other gradients at position 2 to 5 place and to the left in four additional other gradient structure together at position 7 to 10 place Become the gradient profile at isolated edge.Show interpolated peak 3270' substantially at position 5.85, and under interpolated peak 3270' Show vertical axis of symmetry 3271' in dash-dot line.Although each gradient in gradient profile is under the reflection along axis of symmetry 3271' It is not mapped to another gradient in same gradient profile, but it is mapped to interpolated gradient really.For example, at position 3 To the interpolated gradient (substantially at position 8.6) with " X " labelling, it is clearly between position 8 and 9 gradient to gradient map Good interpolation, and vice versa.Equally, the gradient map at position 7 to "+" the interpolated gradient of labelling is (substantially in position Put at 4.6), it is clearly the good interpolation between position 4 and 5 gradient, and vice versa.In addition it can be seen that interpolated Gradient map is to another interpolated gradient, and vice versa, for example, punish in graded levels 0.5 (and respectively at position 4 and 7.6) Not Yong triangle and del labelling a pair interpolated gradient.When gradient profile is interpolated into interpolated gradient profile, Left and right reflection symmetry is especially notable (solid-line curve in Figure 24 J).It shall yet further be noted that on any two graded levels, gradient Midpoint (such as midpoint 3281' and 3280') between the both sides of profile overlaps at same point midway 3287, and this position is also The position at interpolated peak and the position of axis of symmetry.

Be generally understood that is：If any left and right reflection symmetry, along each geometry of vertical axis of symmetry reflection it is then It is mapped to corresponding geometry, overlap with the latter.This kind of geometry include point (that is, gradient or interpolated gradient), Line, curve, there are battery limit (BL), angle etc..Corresponding to this kind of geometry is geometric parameter, for example distance, length, area, inclination angle (that is, the first derivative), curvature (that is, flection and higher derivative) etc..For example, along the area of the profile of interpolated gradient Another section that section (or connecting the linear section of neighbouring gradient) will be mapped on identical graded levels, therefore has identical length Degree.Another characteristic of left and right reflection symmetry is that the point on axis of symmetry is mapped to its own, does not move.Therefore, this grade is put Left, to the distance on the right side be equal.It is also known that being mapped under symmetry between two geometries each other, vertical symmetry Line is placed exactly in the middle of it.Commonly known other characteristics using these characteristics and with regard to reflection symmetry, can design Multiple methods are testing whether gradient profile has left and right reflection symmetry.

Because the property of interpolation is method of approximation, therefore using the pre- constant volume in the parameter of the coincidence for testing geometry Limit area to check symmetry.For example, if X and Y is two parameters measuring, and perfect symmetry will just be produced Raw X=Y, then may specify tolerance limit area so that A<X–Y<B, wherein A and B are positive numbers so that being more than-A and the X Y less than B Value does not lead to the determination of unsymmetry, and the value of the X Y bearing just or than-A than B will lead to excessive lack symmetric determination.Than- A bears must less and the scope of the few value of X Y more positive than B is hereinafter referred to as tolerance limit area, and the boundary in tolerance limit area is unsymmetry Threshold value.Therefore, in this example ,-A and B is the not right of the boundary in tolerance limit area of unsymmetry delimiting X and Y measurement Title property threshold value.When outside the value of X-Y being located at tolerance limit area, hereinafter referred to as exceed (related) unsymmetry threshold value.For example, such as Really (X Y) than B more just, then claims (X Y) to exceed unsymmetry threshold value B.If (X Y) bears than-A, then claim (X Y) to exceed not right Title property threshold value-A.In this context, reception and registration following meanings " are exceeded "：Parameter (X Y) has edge closing due to cross tolerance limit area Boundary and leave tolerance limit area.

Or, can be defined using the function (wherein Z=(1 X/Y)) of Z and reduce compared with making other edges of contributing it The relative extent of the lower contribution to focus signal for this edge.For example, this function of Z can be the sigmoid function of absolute value Z, So that when Z be zero (i.e. X=Y, such as under ideal symmetrical) be this function value one, and with | Z | close to a boundary value Xu Slowly raise, then with | Z | closer to and finally cross this boundary value and rapidly raise, then converge on zero.

It is described below and detects that the gradient profile of bounding edge lacks multiple methods of left and right reflection symmetry, but this is not Exhaustive list.(that is, eliminating or weaken the contribution to focus signal or to focus control for the edge) in this respect of the present invention is not limited to It is discussed below for detecting that the gradient profile of bounding edge lacks the ad hoc approach of reflection symmetry, but it is equal to include it Etc. in method, approximate representation, obvious or known change and the characteristic including using reflection symmetry discussed herein above One or more of any computational methods.

A kind of method in order to detect symmetric shortage be the counting of pixel on the side find out peak gradient with another Difference between the counting of the pixel on side, wherein, the gradient magnitude being associated with pixel is higher than a certain point of peak gradient magnitude Number.(such as) 0.5 can be set as counting unsymmetry threshold value so that when the counting of the pixel of either side has more than opposite side When number exceedes this counting unsymmetry threshold value, symmetric shortage is just detected.By means of Figure 23 A and Figure 23 B and fraction 0.3 And count the example of unsymmetry threshold value 0.5 this situation to be described.In Figure 23 A, posivtive spike for left side and in position 6, And be higher than graded levels+0.3, two pixels are located at left side at respective position 4 and 5, and another two pixel is located at right side each From position 7 and 8 at.Pixel counts are equal between left side and right side, are therefore not detected by unsymmetry.On the other hand, exist In Figure 23 A, for posivtive spike at position 6 for the left side, and it is higher than graded levels+0.3, two pixels are located at left side in respective position Put at 4 and 5, but only one pixel is located at right side at position 7.In the positive gradient profile 3213 in left side in Figure 23 B, left side ratio Right side has more a pixel, exceedes counting unsymmetry threshold value 0.5, symmetric shortage is therefore detected.

A kind of modification is to carry out interpolation to find out the location of pixels 3272 (" interpolated peak position ") of fraction from gradient, wherein Interpolated gradient profile obtains maximum magnitude (" interpolated peak gradient ").This interpolated peak position may be used to calculate to left side, arrives Two distances on right side, as mentioned below.Interpolated peak gradient also may be used to calculate to measure those distances on what graded levels Or on or below what graded levels, pixel is counted.For example, in Figure 24 A, below interpolated peak 3270 Depict vertical chain-dotted line, graded levels 0.95 (upper Grads threshold) streak interpolated gradient profile (in solid-line curve) and draws Horizontal dotted line 3275, streaks interpolated gradient profile on graded levels 0.25 (Gradient threshold value) and depicts horizontal dotted line 3273, in the left side of peak gradient 3212 (in position 6), count two pixels at position 4 and 5 respectively, and existed after right side Only one pixel has been counted at position 7.No matter being reference peak gradient 3212 (in this example at position 6) also with reference to warp Interpolation peak 3270 (substantially at position 5.8), all it doesn't matter with the determination of unsymmetry, because in the former, in counting Peak gradient 3212 is not counted the counting on left side or right side, and in the latter, upper threshold value 3275 (in this example for 0.95, and Typically should be between 0.85 and 0.97) do not allow peak gradient 3212 in the counting counting right side.

A kind of to said method be revised as determine from peak gradient individually to the left, to the right to gradient profile interpolated so that The distance in (" cross point ") in place of crossing over a certain graded levels, this graded levels be peak gradient 3212 (or, interpolated peak 3270) fraction (being preferably between 10% and 90%, be more preferably between 20% and 80%) of Grad, then such as The smaller distance of fruit relatively large distance exceeds more than a certain width unsymmetry threshold value, then just find to lack symmetry.In other words Say, it is more negative than the negative value of width unsymmetry threshold value or than the corrigendum of this width unsymmetry threshold value that distance deducts another distance To lead to determine and lack symmetry.Tolerance limit area therefore occupies the figure blank being symmetrical in zero.Can come one of in a number of ways Determine this width unsymmetry threshold value.Width unsymmetry threshold value can as an image stationary digital or depending on The border width at edge that this peak is associated numeral (10% (if border width is 3 or less) of such as border width, with And 7% (if border width is wider than 3 but being less than 5) of border width) and be given.Gradient wheel (is therefrom produced based on picture signal Gradient in exterior feature) and/or gradient profile in the other rational dependences how to produce of gradient to determining width unsymmetry It is acceptable for threshold value, particularly allow the inaccuracy in interpolated peak and/or cross point.Figure 24 A also illustrates that this is not right Title property detection method.From peak gradient 3212 (at position 6), or can measure from interpolated peak 3270 (at substantially in position 5.8) Distance.In this example, from interpolated peak 3270 measurement distance W_LAnd W_R, sets forth about 2.5 and 1.3, and give 1.2 Difference.Normalization graded levels+0.3 measure border width, gives about 3.7.Can using width unsymmetry threshold value as The fraction of this border width being given, such as 15%, that is, provide 0.56.Because this difference (1.2) exceedes width unsymmetry threshold value (0.56), consequently found that lacking symmetry.It should be noted that no matter being peak gradient 3212 (in this example in position 6 for reference Place) or interpolated peak 3270 (substantially at position 5.8), all not can not ignore with focus signal or with focus control system Relation because the difference between its position typically smaller than 0.5 and often less than 0.3 so that | W_L–W_R| in difference be less than 0.6. It is changed in asymmetric most of situation in the gradient profile on edge, W_LWith W_RBetween mismatch far away more than 0.6.Therefore, One or another one is selected to measure W_LAnd W_ROnly affect the discovery of the symmetric shortage at insignificant minority edge.

A kind of alternative is two areas of estimation, and an area is on a left side for peak gradient 3212 (or interpolated peak 3270) Side, another area on the right side of peak gradient 3212 (or interpolated peak 3270), and according to specified value by this two areas with Area unsymmetry threshold value is compared.Each of two areas can on side by peak gradient (or interpolated peak) below Vertical line delimit, to be determined by multiple interpolated gradients (solid-line curve) (or, connect the straight line of continuous gradient) on another side Boundary, and to be delimited by upper graded levels and Gradient level from top with from bottom, graded levels and this Gradient level on this Different predetermined scores each in peak graded levels (or, interpolated peak graded levels, i.e. the graded levels at interpolated peak) Upper (or, do not go up graded levels to limit this area, but only this face is limited by gradient or interpolated gradient profile Long-pending).For example, in Figure 24 B, wherein gradient profile through normalization so that interpolated peak has graded levels 1.0, Depict graded levels 3276 on 0.75, and depict Gradient horizontal 3274 on 0.2.One area 3277 (has area A_L) (in the left side at interpolated peak 3270) is up delimited by upper graded levels 3276, in lower section by horizontal 3274 demarcation of Gradient, Delimited by the vertical chain-dotted line under interpolated peak on right side, and delimited by interpolated gradient profile (solid-line curve) in left side.Another Area 3278 (has area A_R) (right side at same peak 3270) similarly up delimit with lower section, and on right side by interpolated ladder Degree profile is delimited, and is delimited by vertical chain-dotted line in left side.As area A_LAnd A_RDiffer by more than the predetermined threshold according to specified value When, then shortage symmetry is detected.For example, reach more than area unsymmetry threshold value when larger area has exceeded compared with small area When, then unsymmetry is detected.Can be one of in a variety of different ways expressing this area unsymmetry threshold value.Area is not right Title property threshold value can be expressed according to (compared with small area) percentage ratio, and it can be the stationary digital to an image, or with associated The border width at edge and become.Or, area unsymmetry threshold value can be expressed according to the difference in areas through gradient profile of standardizing. The other how being produced based on the gradient in picture signal (therefrom producing the gradient in gradient profile) and/or gradient profile is reasonable Dependence to determine area unsymmetry threshold value for be acceptable.

In other method shown in Figure 24 I, measurement is from interpolated peak 3270 (or peak gradient 3212) to gradient wheel Wide left side and to right side common apart from W_o.In other words, calculate interpolated gradient (or finding out gradient) so that it is away from through interior Dashdotted distance under slotting peak 3270 (or peak gradient 3212) is W_o.For full symmetric gradient profile, two through interior Slotting gradient is by all on common graded levels.However, for asymmetric gradient profile, two interpolated gradients are located at Different graded levels G_l3252、G_hOn 3253.As graded levels G_lAnd G_hDiffer by more than the predetermined threshold according to specified value When, then shortage symmetry is detected.For example, as larger graded levels G_h3253 are more than less graded levels G_l3252 Amount G_hlExceed less graded levels G_lWhen reaching more than gradient unsymmetry threshold value, can detect that unsymmetry.Can be with various differences One of mode is expressing gradient unsymmetry threshold value.Gradient unsymmetry threshold value can be according to (for example, less gradient G_l) Expressing, it can be the numeral that an image is fixed to percentage ratio, or the border width with associated edge and become.Or, Gradient unsymmetry threshold value can be expressed according to the graded levels' difference through gradient profile of standardizing.(therefrom produced based on picture signal Gradient in raw gradient profile) and/or gradient profile in the other rationally dependences how to produce of gradient for determining gradient It is acceptable for unsymmetry threshold value.Can be by this common W_oIt is chosen as the predetermined score of border width, such as between 0.1 And between 0.5, the fraction that is preferably between 0.2 and 0.4.Or, can be by W_oIt is chosen as from interpolated peak 3270 (or peak ladder Degree 3212) to a pair gradient on the given graded levels as a predetermined score of peak graded levels or interpolated gradient Smaller in two distances.In the case, G_hIt can be individually the parameter indicating asymmetric degree.Then ladder can be set Degree unsymmetry threshold value is so that work as G_hWhen exceeding threshold value, then shortage unsymmetry is detected.As those skilled in the art can Expected, how to define W_oOther changes with above-described standard are possible.

A kind of being revised as respectively in graded levels W to above method_lAnd W_hOn that two interpolated gradients on It is compared between one or flection.For full symmetric gradient profile, two interpolated gradients will be respectively provided with to be removed Other side identical first derivative and sign and value identical flection outside opposite sign.However, it is right In asymmetric gradient profile, this is typically different to first or even flection of interpolated gradient.When the first derivative When value differs by more than the predetermined threshold according to specified value between two interpolated gradients (or possible gradient), just detect Lack symmetry.For example, when larger first derivative has exceeded less first derivative to reach unsymmetry threshold value on value When above, can detect that unsymmetry.As another example, when by one of flection from another one deduct give super Cross unsymmetry threshold value poor when, can detect that unsymmetry.Can be one of in a variety of different ways expressing unsymmetry Threshold value.Gradient unsymmetry threshold value can be expressed according to (for example, less first/second derivative) percentage ratio, and it can be to one Image and fixing numeral, or the border width with associated edge and become.Gradient profile (is therefrom produced based on picture signal In gradient) and/or gradient profile in the other rationally dependences how to produce of gradient for determining gradient unsymmetry threshold It is acceptable for value.

Using pixel counts or distance (Figure 24 A) or area (Figure 24 B) rather than based on interpolated peak or peak gradient Lai Divide count or distance or area, detecting in the said method of unsymmetry, can be using midpoint as previously described.Citing comes Say, using the vertical chain-dotted line immediately below interpolated peak 3270 by between upper graded levels 3276 and Gradient horizontal 3274 Area be respectively divided into area A_LAnd A_RLiang Ge area (referring to Figure 24 B) in, can from upper graded levels (normalization graded levels Horizontal dotted line under 0.75) the middle point-rendering between a pair of cross point between interpolated gradient profile (solid-line curve) is vertical Line.Or, can midpoint between a pair of cross point in Gradient level (normalization graded levels 0.2 horizontal dotted line) Draw this vertical line.Or, can in the range of the predetermined percentage with the graded levels of peak gradient for the interpolated gradient profile (for example Between between 10% and 90%) any graded levels between a pair of cross point between this vertical line of middle point-rendering.

Another kind of modification to above-mentioned area method is estimation above between graded levels 3276 and Gradient horizontal 3274 The length of interpolated gradient curve (or, connect the linear section of continuous gradient).For example, in Figure 24 C, interpolated ladder Line of writing music has the section between normalization graded levels 0.25 and 0.75 in left side and (has length L_L), this section is longer than the right side The section of side, length L of the section on right side_RAppreciably shorter, instruction lacks symmetry.When length L_LAnd L_RDiffer by more than according to finger During the predetermined threshold of calibration standard, shortage symmetry is just detected.For example, reach when longer length has exceeded shorter length When more than length unsymmetry threshold value, unsymmetry is just detected.As area unsymmetry threshold value above, can use multiple differences One of mode is expressing this length unsymmetry threshold value.Length unsymmetry threshold value can be according to (for example, smaller length) Expressing, it can be the numeral that an image is fixed to percentage ratio (preferably from 10% to 30%), or with associated edge Border width and become.Or, length unsymmetry threshold value can be expressed according to the length difference through gradient profile of standardizing.It is based on The other rationally dependences how picture signal (therefrom produce gradient) and/or gradient produce are to determining length unsymmetry threshold value For be acceptable.As described above it should be noted that no matter upper in figure reference peak gradient 3212 is also with reference to interpolated peak 3270, to draw Gradient level or upper graded levels, for focus control system or specifically for focus signal all It is not important.This selection leads to very little difference in length difference, and result is at two edges with relative sign too In so that affecting most of situations of its border width, L_LWith L_RBetween mismatch much bigger.Therefore, select one still Another one is measuring L_LAnd L_ROnly affect the discovery of the symmetric shortage at insignificant minority edge.

Can change and just describe above and the method using Figure 24 C explanation.With its comparison length L_LWith L_R, available similar fashion The respective area A' in the shadow region relatively on left side and right side_LAnd A'_R.Area symmetry threshold value can be similarly defined, and by its For with A'_LWith A'_RBetween the value of difference be compared.

The length method just having described above and having been illustrated using Figure 24 C can be in yet another form changing.By interpolated gradient The curve left side distance (W and upper gradient and Gradient intersection between respectively_BL) and right side distance (W_BR) be compared.If According to specified value W_BLWith W_BRDifference is excessive, then find to lack symmetry.For example, work as W_BLAnd W_BRIn the greater surpass When having crossed smaller and reaching more than width unsymmetry threshold value, find to lack symmetry.And it is possible to it is similar various asymmetric above The mode of any one of property threshold value carrys out specified width, which width unsymmetry threshold value.

Figure 24 H illustrates a kind of alternative method, and it is equivalent to the width method of just description above.The method calculates upper gradient water The first midpoint 3281 (upper midpoint) between two interpolated gradients 3284,3285 under flat 3276, and Gradient level The second midpoint 3280 (lower midpoint) between two under 3274 interpolated gradients 3282,3284, then calculates the first midpoint 3281 and second the distance between midpoint 3280 X_ab.Full symmetric gradient profile is at the first midpoint 3281 and the second midpoint 3280 The distance between (" distance between midpoint ") be insignificant, that is, midpoint overlap, as in Figure 24 J.When between midpoint away from From when exceeding a certain middle dot spacing from unsymmetry threshold value, shortage symmetry is just detected.Distance between midpoint is | W_BL–W_BR| Twice.In a kind of change of the method, simply use graded levels 3274, and only calculate corresponding midpoint 3280, then measure from peak gradient 3212 (or, interpolated peak 3270) to independent midpoint 3280 apart from X_bPk, that is, as Figure 24 H In also show like that.

Another method is the slope of the gradient profile on the both sides find out peak gradient 3212 (or, interpolated peak 3270) The first derivative, and this is compared to the first derivative according to specified value, to determine whether to lack symmetry.For example, if Compared with the value of the first derivative that the value of the first derivative of steep slope exceedes relatively slow slope reach the first derivative unsymmetry threshold value with On, then just find to lack symmetry.On the other hand, full symmetric gradient profile will have the of same magnitude on both sides One derivative, it is only different in sign.First derivative can generally be calculated by interpolation.Can be as peak Grad The gradient water of a certain fraction (being preferably between 10% and 90%, be more preferably between 20% and 80%) of (such as 0.5) The first derivative is generally calculated on flat.In Figure 23 B, it is 0.5 times of part of the graded levels of peak gradient 3212 in graded levels, There is the slope continuously raising or declining a pair, every side of posivtive spike gradient 3212 has one.The slope on right side is significantly More precipitous than the slope in left side.Figure 24 D shows how to estimate the first derivative to normalization gradient profile on graded levels 0.25, and Made approximately with the hypotenuse of right angled triangle (shade), this is 1 to the base width of right angled triangle, and respectively contacts through interior The left side of slotting gradient profile and right side.This is to the first derivative respectively by triangle S_LAnd S_RHeight approximate to make.Figure 24 E explanation Using two trianglees with identical base width (being 1 in here explanation), to the first derivative, approximate the opposing party is made to this Formula.The base of each triangle is on graded levels 0.25, the interpolation position of gradient profile is placed in the middle.Every one end of corresponding hypotenuse Away from half-pixel, and take from the Grad of gradient profile interpolation as Grad.First derivative is approximately height S'_LWith S'_RDivided by base width, it is 1 in the explanation of base width here.As is generally known, exist and from sequence of data points, the first derivative is made Approximate method have several, therefore the present invention be not limited to particular instance given above in this respect, but include in order to All equivalent methods of the first derivative and all method of approximations are provided.

Another method is second of the gradient profile on the both sides find out peak gradient 3212 (or, interpolated peak 3270) Derivative, and this is compared to flection according to specified value, to determine whether to lack symmetry.For example, if one surpasses Cross another one and reach more than the unsymmetry threshold value of flection, then just find to lack symmetry.Conversely, full symmetric gradient Profile will have identical flection on both sides, its sign and value all same.Generally can be calculated by interpolation Flection.(can be preferably between 10% and 90%, be more preferably situated between in a certain fraction as peak Grad (such as 0.25) Between 20% and 80%) graded levels on calculate flection.Figure 24 F explanation can how approximate flection method. First, the equally spaced gradient of continuous phase of tlv triple in left side or interpolated gradient 3295,3296,3297 and right side are found out Another tlv triple 3291,3292,3293, wherein respective midpoint 3296,3292 is on normalization graded levels 0.25.For three Given interval in tuple, finds out a normalization to (3295, the 3297 of left side and 3291, the 3293 of right side) for other external point The arithmetic mean of instantaneous value of Grad its be higher than normalization graded levels 0.25 height (D_LAnd D_R), to lead as to respective second Count is approximate.As shown in Figure 24 F shows, D_LFor just, and D_RIt is negative.

Figure 24 G illustrates another kind of method of approximation of flection on normalization graded levels 0.18,.On each side, one Individual triangle is coupled to the gradient profile above it, and another triangle is then in interpolated gradient profile and graded levels 0.18 Below (that is, the gradient under this graded levels or interpolated gradient) both cross points.Each hypotenuse is with an inclination angle Tilt, so that hypotenuse is coupled to this interpolated gradient profile.The height S of left side triangle_ULAnd S_LL(the height of right side triangle S_URAnd S_LR) subtract each other to draw flection D_L(D_R).Arrow indicates this relative sign to flection：D'_LRefer to upwards, D'_RDownward finger.Flection on this gradient profile both sides significantly notable mismatch, is therefore asymmetric.

As is generally known, carrying out approximate flection from sequence of data points has multiple methods, therefore the present invention in this respect not It is limited to particular instance given above, but include all equivalent methods for flection and possible approximate.

Although above example description detect shortage reflection symmetry using different measuring parameter, can use simultaneously with Its other measurement parameter is calculated, then these measurement parameters are combined to form at comparing ginseng more than both in upper method Number, with unsymmetry threshold value or be compared with predetermined tolerance limit area in a broad sense so that when compare parameter value in tolerance limit area Outside when, shortage reflection symmetry is just detected.For example, the distance between midpoint X_ab(Figure 24 H) and the first derivative it Between poor S_R–S_L(Figure 24 D) can be combined in weight averaged value Z, is then compared with unsymmetry threshold alpha, and it is by tolerance limit Area definition is interval α<Z<α.

Detect that having unsymmetry in this edge gradient sequence on themselves is not necessarily related to directly to this gradient Sequence calculates.

Unsymmetry can detect in the multiple images sample value on crossing over this edge.Referring to Figure 26, for edge Be respectively provided with contrast C 1 and two non-branches the narrowest of C2 divide, if having ideal symmetrical in multiple gradient such as this, then Both centers will be mated.Misalignment between Liang Ge center indicates unsymmetry.Can measure this misalignment and by its divided by The width at edge, to provide the amount of the unsymmetry in the gradient profile that indicate on this edge.

Or, from the center that the non-branch corresponding to contrast C 1 divides to this gradient sequence have rising the most precipitous or At fall, (the peak gradient corresponding to Robert's detector is g_peak) distance, and divided by edge width to indicate asymmetric journey Degree.

Further alternatively, unsymmetry can be measured from multiple flections of multiple images sample.Referring to Figure 25, bear It is asymmetric that the difference of the peak and posivtive spike distance of interpolated zero passage between both indicates one divided by distance between the two Degree.

Above-mentioned detection and the solution party for asymmetric edge can be executed in rim detection with width measure unit 206 Case.

Note, in the present invention, when the amount (such as graded levels) from edge is divided by (acquiescence, unless otherwise specified) side During the Grad at the peak Grad of edge or interpolated peak, this amount is referred to as to standardize.For example, in Figure 23 B, peak gradient 3212 have the just normalized value for 1, and in Figure 24 C, interpolated peak 3270 is different from peak gradient 3212, and Figure 24 C Shown gradient is standardized with respect to interpolated peak 3270 rather than peak gradient 3212.

Lengthy filter

The function of lengthy filter 212 is described below.Inventionbroadly, lengthy filter 212 forms to being connected respectively to class Preference like the edge at one or more edges of orientation.It is similarly oriented in group and interconnective edge group (" level Connection edge ") compared with the isolation edge not contacting any other edge being similarly oriented, it is less likely to be caused by noise.Therefore, The edge being similarly oriented is more for having of being concatenated together, and the probability that it is caused by noise is less.This group is caused by noise Probability increase and exponentially decline with the number at the edge in group, and far faster than linear mode.Can be using this property Giving up noise, especially under dim light or short exposure situation, wherein signal to noise ratio is weaker for matter, and e.g., less than 10, in image Interior or in regions of interest.Preference can be implemented to express this preference in any rational method.Some modes described below It is only example.

First method is to eliminate to belong to have the side cascading edge less than the vertical/horizontal of the length of cascade length threshold Edge.When regions of interest is dimer, cascade length threshold can be larger.For example, cascade length threshold can start little to 2, but Drop to 5 and increase to 8 with the signal to noise ratio in regions of interest.Cascade length threshold can by processor 112,112', 112 " (for example By " length command " signal shown in Fig. 3) provide as a part for signal 132.Or, can be produced according to focus signal Formula on device is calculating threshold value.

Second method is to be directed to each edge in lengthy filter 212 to provide a Length Weight, and in focus signal This Length Weight is applied to calculate focus signal in computer 210.Compared with the edge of the part as shorter cascade edge, As the weight that the marginal reception of the part at longer cascade edge is larger.For example, Length Weight can be cascade edge Length square.Therefore, each edge can be made to be multiplied by factors A/B to the contribution of focus signal, be asked to contributing afterwards With to form focus signal, wherein B is the summation entering the Length Weight at all edges that focus signal calculates, and A is edge Length Weight.Equally, can be able to have as longer cascade side as the border width rectangular histogram of the part output of signal 134 The edge of the member of edge, it is preferred, therefore, that more contribute corresponding to its respective edges width and therefore for preferred area Between, rather than all edges all contribute identical amount, such as+1.So that it takes up a position, for example, A/C, wherein C can be contributed in each edge Meansigma methodss for the A on this edge.Similarly, narrow edge counts the member can having as contributing more longer cascade edge Edge.So that it takes up a position, for example, the contribution from each edge can be multiplied by A/D, wherein D is to count in narrow edge counts The meansigma methodss of the A between edge.

Group's (wherein, top (the most left) and bottom (the rightest) edge except) at N number of vertical (level) edge, each edge Vertical (level) edge of two other of contact, an other edge (left side) above edge itself, and under this edge itself Vertical (level) that side (right side) another other edge is N for length cascades edge.Top (the most left) edge only needs to touch An edge on (right side) below itself.Bottom (the rightest) edge only needs to touch an edge in (left side) above it.

Fig. 8 explanation is vertical to cascade edge and its length.In fig. 8, unit R 2C3 and R2C4 form the first vertical edge, single First R3C3, R3C4 and R3C5 form the second vertical edge together, and unit R 4C4 forms the 3rd vertical edge together with R4C5. First and the 3rd vertical edge each only other vertical edge of touching one, and the second vertical edge touches two other vertical edges Edge.First, second, and third vertical edge forms the vertical cascade edge with length 3 together.

At vertical (level) cascade edge, there are two or more branches (that is, having two edges in row (column)) Situation (not shown) in, length can be defined as cascading the sum at intramarginal edge.Or, length can be defined as from it In top (the most left) edge vertical (level) distance of adding one to bottommost therein (the rightest) edge.

In addition to above proposal, other possible methods are had to define cascade length.For example, cascade the length at edge Define and should have the following properties that：Length is proportional to the number at the member edge of at least up to three in cascade edge.This will be with The inference of prior statement is consistent, and this inference is to be connected with each other by contacting with each other compared with multiple edge exponentially to be reduced by making an uproar The probability at the cascade edge that sound causes, and therefore length should be expressed and cascade the number at the member edge reaching reasonable number in edge Proportionality, this reasonable number fully strengthen cascade edge confidence level, past the confidence level of single member.Length is filtered Ripple device 212 can mitigate or exclude and therefore (in general) repel repel cascade length be one edge.Lengthy filter 212 can Repel the edge that cascade length is two.Lengthy filter 212 can repel the edge that cascade length is three, to reduce noise further Impact.Lengthy filter 212 can carry out any one of these actions according to the order from processor.

Although being shown as in Fig. 3 after rim detection with width measure unit 206, other arrangements are possible 's.For example, can before focus signal computer 210 intubating length wave filter 212, wherein by lengthy filter 212 The edge of reason is the edge through spanning filter 209, and it depends on " refinement " signal.

In the alternate embodiment of focus signal generator, removable refinement switch 220 is so that focus signal computing unit 210 receive the first group of data not filtered and filtered second group of data by spanning filter 209, and for each meter Calculate different focus signals, calculate a mao focus signal for the former, calculate fine focusing signal for the latter, and by above two signal Export processor 112,112'.

Spanning filter

With reference next to Fig. 3 to understand the operation of spanning filter 209.Figure 11 describes the rectangular histogram of border width, i.e. side Edge counts the curve chart to border width.At border width 2, that is, there is peak in above-mentioned sharp_edge_width, its Indicate the presence of sharp edges in image.However, at border width 4 and 5, there is peak, its instruction may be because of corresponding imaging Edge outside focus, fuzzy at the different distance away from condenser lenses compared with the object producing sharp edges for the object. In order to calculate focus signal, width side in preset range (" narrow edge scope ") outward can be mitigated using spanning filter 209 Edge.Spanning filter 209 can produce less weight in the extraneous border width of narrow edge, so that focus signal calculates Middle use.For example, weight 1.0 can be assigned for border width, and be that the border width more than+1 is assigned on the upper limit 840 right side Weight 0, and the border width between being assigns the weight between 0 and 1.0, the monotonic decreasing with border width.Or, width Wave filter 209 can be prevented that a little edges fully enter focus signal here and calculate.The suitable upper limit 830 and lower limit 840 are depending on some Factor, including the crosstalk in pel array 108, the missing color in order to produce the image being received by focus signal generator 120 Interpolating method, and for the wave filter system in the low pass filter used in rim detection and width measure unit 206 Number.By catching the image of various acuity and border width rectangular histogram can be checked come true for image pick up equipment 102,102' The fixed suitable upper limit 830 and lower limit 840 and parameter sharp_edge_width.For example, if sharp image is at edge At width 2, there is peak, then suitable lower limit and the upper limit can be respectively 1.5 and 3, and can be set as sharp_edge_width 2.0.Lower limit and the upper limit and sharp_edge_width can be determined as above, and pass through processor 112,112 " it is supplied to focusing letter Number generator 120,120', 120 ".When " refinement order " is for ON, the fine focusing signal so calculating mitigates in narrow edge scope Outer border width.

In addition, spanning filter 209 can calculate border width falling the grand total at the edge in the range of narrow edge, and conduct A part for output signal 134 and export.Narrow edge can be counted and be input to focusing system controller (processor 112) and by gathering Burnt system controller (processor 112) uses, to detect the presence of sharp image and/or to be used for initial tracking.

Focus signal

With reference next to the focus signal computer 210 of Fig. 3, focus signal computer 210 receives border width and output Focus signal.Focus signal can be calculated as the weight averaged value of all border widths, wherein weight is for each edge The edge of width counts, i.e. focus signal=Σ w_ie_i/Σw_i, wherein e_iFor border width, w_iFor weight, wherein w herein_i= c_i, c_iFor border width e_iThe edge at place counts, and i is the histogrammic interval number of border width.Or, each border width The weight at place can be that the edge counting for border width is multiplied by border width itself, i.e. w_i=c_ie_i.In addition, from width filter The preference expressed according to weight of ripple device 209 can be multiplied by each border width further.For example, for spanning filter Weight Ω produced by 209_i, Σ Ω_i=1, focus signal can be calculated as Σ Ω_iw_ie_i/ΣΩ_iw_i.If control signal is " thin Change " it is ON, and " exclusion " is OFF, then focus signal is by for the sharp edges width very close to the example shown in Figure 11 2.0 value, thus instruction will produce in the border width between 2.0 and 3.0 object detail in the range of focusing distance Among, most of object detail are practically at sharp focusing.If control signal " refinement " is OFF, and " exclusion " is OFF, So focus signal can be the value near 5.0, thus indicating the most of details that there is the image outside focus.Connect refinement to open Close 220 permission focus signals and more respond slightly fuzzy object, and less respond completely fuzzy object.When refinement is opened When pass 220 is ON, focus signal will be referred to as fine focusing signal, and when refining switch 220 for OFF, focus signal will be claimed For hair focus signal.As previously mentioned, can be in one of several ways in which mode by emphasizing expressed by lengthy filter 212 It is incorporated in focus signal, for example eliminating the edge being mitigated makes it not enter focus signal calculating, or reduces edge to correspondence The interval counting e of border width_iContribution weight.

Figure 15 sketches the response that fine focusing signal is in the adjustment of the focal position of proximate of sharp focusing to object. Fine focusing signal reaches minima, and approximately at sharp_edge_width, wherein focal position makes image reach sharp focusing, And otherwise just increase.Fine focusing signal can be used for following the tracks of the object focusing on or being in close proximity to focusing.For mobile object, Fine focusing signal allows focus control system to make object keep sharp focusing, even if it is also such that focusing distance constantly changes. Fine focusing signal can be additionally used in obtaining be not yet in sharply focusing on but the sharp focusing (" acquisition ") of object that sufficiently closes to so that This object produces width and falls the edge in the range of narrow edge.Because border width rectangular histogram is corresponding to away from sharp_ Peak is represented at the border width of the object of edge_width, thus leading to fine focusing signal to be more than sharp_edge_width, because This focus control system can be responded by following action：Adjustment focal position is so that fine focusing signal value is towards sharp_ Edge_width, thus because the object at edge width values makes the peak of border width placed in the middle equal to sharp_edge_width.

Basic use

How Figure 12 to Figure 16 explanation can be counted using narrow edge, hair focus signal and fine focusing signal execute focusing control System is to realize sharp image.

Figure 12 explanation has the Outdoor Scene of 3 object groups at different focusing distances：" people " in the foreground, " mountain range, the sun and horizon " in background, and " car " in-between.

Figure 13 is when the focal position of condenser lenses 104 is from scene of illustrated in fig. 12 swept as far as near-earth, right The explanation that the narrow edge described according to the time counts.When focal position makes object reach the sharp image on pel array 108, Narrow edge counts and reaches peak.Therefore, narrow edge count curve represents 3 peaks, during scanning, " mountain range, the sun and horizon ", " automobile " and " people " is with each peak of this order.

The hair focus signal that Figure 14 shows the comparison time and describes.Reach the 3 of peak when focal position counts close to narrow edge During each of individual focal position, hair focus signal represents minima.However, at each minima, hair focus signal is simultaneously It is not at sharp edges width levels, it is 2.0 in this example, because the edge width that the other objects outside focus are contributed Degree is larger.

Figure 15 explanation is described to the focal position near the sharp focal position of " automobile " in the scene impinging upon Figure 12 Fine focusing signal.Fine focusing signal substantially realizes sharp edges width, and it is 2 in this example, despite the presence of fuzzy right As (" people " and " mountain range, the sun and horizon ").Referring again to Figure 11, wherein two peaks at width 4 and 5 are by described two moulds The contribution of paste object group, this can be regarded as spanning filter 324 and has reduced weight or be completely eliminated the comfortable upper limit 840 right side The contribution of border width.

Focus control system can search for nearest sharp focal position in search pattern using hair focus signal.Focus on Focal position can be moved apart present convergence location by control system, to determine that a mao focus signal is to increase or reduces.For example, If when focal position is inwardly (outside) mobile, hair focus signal increases (reduction), then exist from present convergence location relatively Remote sharp focal position.Processor 112,112', 112 " can then provide focus drive signal, with towards neighbouring sharp Mobile focusing lens 104 on the direction of focal position.

Focus control system can be followed the tracks of using fine focusing signal already at the sharp object focusing on, so that corresponding figure As maintaining sharp (therefore " tracing mode "), but regardless of the movement of the change of scene, the movement of object or image pick up equipment.When When object is in sharp focusing, fine focusing signal level is stable, but regardless of this changes.Therefore, the change of fine focusing signal Change the change of the focusing distance to image distance image pick up equipment for the hint.Near by focus control system " locked " to minima Given fine focusing signal level, for example in this example between 2.0 and 2.5 it is manifestly that 2.1, fine focusing signal electricity Flat any displacement can inform immediately processor 112,112', 112 " change of the focusing distance of object.Processor 112,112', 112 " then can determine direction, and cause condenser lenses 104 mobile so that fine focusing signal level returns to " locking " level.Cause This, image pick up equipment 102,103,103', 103 " mobile object can be followed the tracks of.

Focus control system (for example processor 112,112', 112 " in algorithm in implement) can be counted using narrow edge To trigger the change from search pattern to tracing mode.In the tracking mode, focus control system carrys out " lock using fine focusing signal Fixed " object.Before focal position is sufficiently close together the sharp focal position of object, focus control system can use hair focus signal To identify moving direction and to adjust the translational speed of lens.When object enters sharp focusing, narrow edge counts and sharp reaches Peak.Detect narrow edge counting drastically rising or reach peak or both after, processor 112,112', 112 " can at once be switched to In tracing mode, and it is focused position control using fine focusing signal.Can (each different sharp focal positions can by threshold value Different) it is assigned to each object group finding from end-to-end focal position " scanning ", and subsequently exceed this when narrow edge counts During threshold value, detect corresponding object group.For static scene, for example, still image is shot, end-to-end focal position is swept Retouch and can return to maximum count list, each peak that reaches that narrow edge counts has a maximum count.Can (such as) by taking maximum Count 50% and from maximum count list produce threshold list.

Figure 16 illustrates image pick up equipment 102, and it has display 114, includes the input equipment 107, Yi Ji of button The selected marker 1920 highlighting in display 114.User can be created using input equipment 107, shape and manipulate selected marker 1920.Although being shown as including button in this example, input equipment 107 may include the touch screen overlying on display 114, To detect touch or beating position on display 114.Input equipment 107 and processor 112,112', 112 " or be used for inputting The single nonshared control unit (not shown) of device 107 can determine that selection area.(or can produce in focus signal via bus 132 In the case that device 120 is a part for processor 112, inside processor 112) parameter being used for describing selection area is transmitted into Focus signal generator 120,120', 120 ".As response, focus signal can be calculated or narrow side by focus signal generator 120 Edge count or both be limited to by described parameter describe select area in edge, or mitigate select area outside edge.Do so can Mitigate non-the wanted object from focus signal, and then even hair focus signal also will show single minima and sharp The 1.0 of border width or minimum levels interior below.

Alternate embodiment

Figure 17 shows the alternate embodiment of focus signal generator 120'.Focus signal generator 120' output edge and side The statistics of edge width.Can be one of the following or one among the border width statistics of controller 120' output with On：Border width rectangular histogram, its edge including under different border widths counts；Wherein border width counts and reaches maximum Border width；One group of coefficient, it represents the spline function estimating that the edge under different border widths counts；And edge can be represented Any data of the function of width.GeneraI investigation unit 240 can receive in other units with focus signal generator 120' The statistics to calculate border width for the data of one or more middle calculating.In general, focus signal generator 120' can Output has the signal of the instruction of distribution of border width.

Referring to Figure 18, to the processor 112' substituting in automatic focusedimage pick device 102' therefore in signal 134 The border width statistics that provides of alternate embodiment can be used for according to method discussed herein above by processor 112' or it is equivalent Thing counts calculating hair and/or fine focusing signal and narrow edge.In addition, can be by calculating in focus signal generator 120' Any data exports processor 112' as a part for output signal 134.

In addition to the function in the processor 112 being included in Fig. 1, processor 112' can internally produce focus signal and/ Or narrow edge counts.

Pel array 108, A/D converter 110, color interpolation device 148 and generator 120' can reside within encapsulation 142, Constitute the imageing sensor 150' separating with processor 112' together.

Auxiliary pixel array

Figure 19 shows the alternate embodiment of automatic focusedimage picking up system 103.Except be included in element in system 102 it Outward, system 103 may include partially reflecting mirror 2850, completely reflecting mirror 2852, optical low-pass filter 2840, main pel array 2808 And main A/D converter 2810.Incoming beam splitting can be become the first divided beams and the second divided beams by partially reflecting mirror 2850, and one Person is transmitted, and another one is reflected.First divided beams can further pass through light before eventually arriving at main pel array 2808 Learn low pass filter 2840, its detection first divided beams is simultaneously converted into analogue signal.Second divided beams eventually arrive at auxiliary Help pel array 108 " can be reflected by completely reflecting mirror 2852 before, auxiliary pixel array 108 " corresponding to the system 102 shown in Fig. 1 In pel array 108.The ratio of the light intensity of the first light beam and the second light beam can be 1 to 1 or be more than 1 to 1.For example, Ratio can be 4 to 1.

Main pel array 2808 can be covered by the colorful optical filter array of color mosaic chart case (such as Bayer pattern).Optics It is too small and lead to aliasing that low pass filter 2808 prevents from focusing on smallest spot on pel array 2808.Coloured silk in mosaic Colo(u)r filter covers the place of pel array 2808, and aliasing can cause color moire fringes illusion after color interpolation.Citing comes Say, by using optical low-pass filter, 84% circle of the visible power of luminous point on the main pel array 2808 of encirclement Minimum diameter (" minimum full diameter ") can be remained above one and 1/2nd pixel wide, but is less than two pixel wide.Citing For, if main pel array 2808 has the pixel wide of 4.5um, and minimum diameter is 2.0um, and no optics low pass filtered Ripple, then optional optical low-pass filter 2840 is so that a diameter of 6.7um or bigger of luminous point.

Auxiliary pixel array 108 " may include one or more photodetector arrays.Each of described array Can by or can not by color mosaic chart case colorful optical filter array cover.Array in auxiliary pixel array 108 " is in simulation letter Output image in number, described analogue signal is converted into digital signal 130 by A/D converter 110.Send images to focus signal Generator 120.Color interpolation device 148 can produce the missing color of the image producing from the pixel being covered by colored filter.As Fruit auxiliary pixel array 108 " includes multiple photodetector arrays, then every an array can catch corresponding to by main pel array The subimage of a part for 2808 images capturing.The plurality of array can physically separate more than 100 pixel wide, And can share or Semiconductor substrate can not be shared.In auxiliary pixel array 108 " in pel array do not share Semiconductor substrate In the case of, described pel array can be contained in an encapsulation (not shown) together.

The analogue signal of main A/D converter 2810 autonomous pel array 2808 in the future is converted into digital master image files signal 2830, it is sent to processor 112, and the image acceptable images wherein capturing on main pel array 2808 are processed, example As color interpolation, colour correction and image compression/decompression, and it is ultimately stored in memory card 116.

Photodetector array in auxiliary pixel array 108 " can have the pixel wide less than main pel array 2808 The pixel wide (" auxiliary pixel width ") of (" main pixel wide ").Auxiliary pixel width can be as small as the half of main pixel wide. If auxiliary pixel is covered by colored filter, and in the case of no low-pass optical filtering, auxiliary pixel width is less than visible 1.3 times of the smallest spot of light, then can be in auxiliary array 108 " are previously inserted into the second optical low-pass filter, so that auxiliary picture Minimum diameter in pixel array 108 " (" minimum auxiliary diameter ") increase to big at 1.3 times to 2 times between, but main still less than minimum Diameter, preferably 1.5.Slight molar stricture of vagina in assistant images is not a problem, because assistant images are not intended as final seizure figure As presenting to user.

Figure 22 explanation for the master image (solid-line curve) carrying out autonomous pel array 2808 and is derived from auxiliary pixel array 108 " Assistant images (imaginary curve), how border width can change with regard to sharp focal position.Proper master image reaches is calibrated Sharp edges width 2 when, assistant images provide more sharp slope.Permit assistant images to reach less than calibrated sharp edge Edge width, because the moire fringes leading to because of aliasing are in auxiliary pixel and less crucial, because assistant images are not intended as Final image presents to user.This contribute to making sharp edges width above and below slope sharp.More sharp slope Promoted by the auxiliary pixel width less than main pixel wide.

Shadow region instruction in Figure 22 controls focal position in it so that master image keeps the sharp good area focusing on.Poly- The outside change in burnt position will lead to border width to increase in assistant images, and inwardly change and will lead to border width in auxiliary Reduce in image.In order that the border width of master image maintains close to sharp edges width, Linear Feedback Control Systems can be used So that middle assistant edge width value to set the goal in shadow region, and the border width producing from assistant images is used as feedback letter Number.

Auxiliary pixel array 108 ", A/D converter 110, focus signal generator 120 may be housed in encapsulation 142 together, And constitute aiding sensors 150.Aiding sensors 150 can further include color interpolation device 148.

Figure 20 shows and replaces outer oneself similar to equipment 103 of focus signal generator 120 except focus signal generator 120' The alternate embodiment of dynamic focusedimage pick device 103'.Auxiliary pixel array 108 ", A/D converter 110, focus signal produce Device 120 may be housed in encapsulation 142 together, and constitutes aiding sensors 150'.Aiding sensors 150 can further include color Interpolater 148.

The alternate embodiment of Figure 21 automatic focusedimage pick device 103 of displaying ".Focus signal generator 120 and process Device 112 " can be contained in encapsulation 144 as camera controller, with auxiliary pixel array 108 " separate.Processor 112 " similar to Processor 112, difference is processor 112 " receive autonomous pel array 2808 and auxiliary pixel array 108 " Image.Processor 112 " can perform color interpolation, colour correction, compression/de-compression and will receive on signal 2830 Image stores memory card 116, can be to the process of signal 130 execution similar to processor in Fig. 2 112.Different from Fig. 2 It is that this is in the image receive on signal 130 and need not accept compression/de-compression and store memory card 116.Processor 112 " can be for by auxiliary pixel array 108 " in the pixel that covers of colored filter to the image receiving on signal 130 Execution color interpolation, and on signal 146, the image through color interpolation is sent to focus signal generator 120.

Automatically focusedimage picking up system 102,102', 103,103', 103 " may include computer program storage medium (not Diagram), it include causing respectively processor 112,112', 112 " and/or focus signal generator 120,120' execute this paper institute One of function of description or one or more of instruction.For example, described instruction can cause processor 112 or generator 120' corrects to border width execution gradient according to the flow chart of Fig. 7.As another example, described instruction can cause processor 112' or generator 120 are according to the description execution border width filtering above with respect to spanning filter 209.Or, processor 112nd, 112' or generator 120,120' may be configured to have for one of function included by this paper or one or more of The combining or pure hardware embodiments of firmware and hardware.For example, in generator 120, can execute in pure hardware and incline Slope correction, and according to the instruction execution lengthy filter 212 in firmware.

Figure 30 shows the another embodiment of focus signal generator 120'.This embodiment can be in images above seizure system Any one in adopt.

Although memory card 116 to be shown as the part of system 102, can be changed to using any non-volatile memories matchmaker Body, such as hard disk drive, the image being wherein stored therein can be accessed by user, and reproducible to system 102 outer and away from The diverse location of system 102.

It is storable in system for one of described system or more than one parameter (such as sharp_edge_width) In nonvolatile memory in interior device.Described device can be flash memory device, processor or imageing sensor, or Focus signal generator as disconnected device.By one of system or more than one formula (for example by based on Calculate cascade length threshold, or be used for calculating β) equally can be stored in those devices as parameter or as computer executable instructions One of or one or more of in nonvolatile memory in.

Although having described that and illustrate some one exemplary embodiment in accompanying drawing, it will be understood that, this little embodiment only illustrate and The unrestricted present invention, and the invention is not restricted to the particular configuration showing and describe and arrangement because the technology people of art Member is contemplated that various other modifications.

Claims (9)

1. a kind of multiple edges for the image from scene produce the method to indicate image acuity for the focus signal, its Including：

In a computing device, based on an amount of the asymmetric degree in the gradient sequence indicating on an edge, mitigate or arrange Remove the contribution to this focus signal from this edge,

Wherein, any edge described focus signal being made to contribute all contributes an edge sharpness metric, and this edge sharpness is measured and is Multiple images data sample in the predetermined neighborhood of described any edge and the amount that calculates,

Wherein:

(A) this edge sharpness measures the width that the predefined non-branch for described any edge divides, and described predefining is not divided Part is that the plurality of view data sample based on described any edge defines in a predefined manner, and：

I () described predefined non-branch is divided into the described arbitrary of the predetermined score of a contrast contributed on described any edge The Zhai Wei branch at edge divides；Or

(ii) described predefined non-branch is divided into a positive peak or interpolated of a second dervative sequence on described edge Kuan Wei branch between peak and a negative peak or interpolated peak divides；Or

(B) this edge sharpness measures a peak Grad for the gradient sequence on described any edge divided by described any edge On a contrast；Or

(C) this edge sharpness measure be a gradient sequence from described any edge multiple gradients to respect to the plurality of Gradient and a function of the distance of a predefined position.

2. method according to claim 1, wherein, in (C) item, using Grad as weight, then described predefined Position is the center of gravity of the plurality of gradient.

3. method according to claim 1, wherein, in (C) item, described function is measurement described predefined position week The kth central moment of the plurality of gradient enclosed, k is a positive even-integral number.

4. the method according to any claim in the claims, it further includes：

Described edge is detected to be associated with the gradient sequence across its own, and one indicates the asymmetric degree in this sequence Measure had value more than an asymmetric threshold value or outside a predetermined tolerance limit area.

5. the method according to any claim in claims 1 to 3, wherein, if it is indicated that this amount of asymmetric degree More than a threshold value, then exclude this edge in this focus signal is contributed or mitigate its weight.

6. the method according to any claim in claims 1 to 3, wherein, using this amount indicating asymmetric degree To reduce the weight in the weighted average that the edge sharpness at multiple edges that this focus signal is made to contribute for this edge is measured.

7. the method according to any claim in claims 1 to 3, wherein, a second dervative from this edge A pair of posivtive spike of sequence to be counted with the distance of a midpoint to an interpolated position of the zero passage of this second dervative sequence of negative peak Calculate this amount indicating asymmetric degree.

8. a kind of circuit, multiple edges of its image from a scene produce a focus signal to indicate the sharp journey of an image Degree, this circuit includes：

Rim detection and width measure (EDWM) component；And,

Focus signal calculates component,

Wherein this rim detection and width measure component detect the edge in the view data of this image, determine for the edge such as this It is made the relative extent contributed respectively to this focus signal, and for the edge assessment edge that will this focus signal is made to contribute Sharpness metric,

Wherein pass through to consider this corresponding relative extent of grade, this focus signal calculates component and gathers from the edge such as this sharpness metric generation one Burnt signal,

Wherein this rim detection and width measure (EDWM) component are implemented according to right arbitrary in above method claim 1 to 7 Require described method.

9. image capture system, it includes：

Condenser lenses；

Imageing sensor, it includes an image sensing pixel array, and this image sensing pixel array receives is gathered by this condenser lens Burnt light thereon；

The motor components of this condenser lens；And,

Circuit as described in above claim 8.