Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/20—Image enhancement or restoration using local operators
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V20/00—Scenes; Scene-specific elements
  • G06V20/10—Terrestrial scenes
  • G06V20/13—Satellite images

Definitions

• the present invention relates to image enhancement, and in particular to a method and apparatus for enhancing an image of a scene partially obscured by atmospheric backscattered light.
• Various techniques are known for enhancing images of scenes which are obscured by light backscattered from, for example, the atmosphere.
• a camera may be mounted on an aircraft to obtain a view of the terrain scene over which that aircraft is flying. Assuming that the scene is obscured by low mist, cloud or other atmospheric effects, the intensity of light reaching the camera from those terrain features contributing to the scene is reduced. A small amount of light scattered from the terrain does reach the camera, but this is obscured by light scattered from the mist or cloud.
• the gain of the camera or other sensing s> stem is set. usually by an automatic gain control, to the maximum brightness of the image.
• the transmitted terrain component becomes small in comparison with the quantisation noise of the sensor.
• the backscattered light often has a random component, and this is a source of noise which is amplified by any contrast-stretching transformation implemented by the sensor.
• DTE Digital Terrain Elevation
• Enhanced images generated using the known motion-compensated averaging systems may be further improved by contrast enhancement.
• contrast enhancement algorithms are known, for example variance normalisation or histogram equalisation. In practice however such known contrast enhancement algorithms have not provided particularly good results.
• a method for producing an enhanced image of a scene partially obscured by backscattered light wherein an image of the scene is captured, a model is defined to represent the relationship between the brightness of a pixel of the image and the distance between the point the view from which the image represents and the point in the scene represented by that pixel, an estimate of the contribution of backscattered light to the brightness of each pixel of the image is computed from the model, the estimated contribution for each pixel is subtracted from the brightness of that pixel to produce a modified brightness for that pixel, and the enhanced image is formed by allocating to each pixel of the image a brightness which is a function ofthe modified brightness of that pixel.
• a series of images of the scene is captured and an averaged image is produced in which each pixel has a brightness which is an average of the brightness of those portions of the captured images that represent the same region of the scene as that pixel, the averaged pixel then being processed.
• a series of images may be processed separately and then averaged.
• the invention also provides an apparatus for producing an enhanced image of a scene partially obscured by backscattered light, comprising means for capturing an image of the scene, means for defining a model to represent the relationship between the brightness of a pixel of the image and the distance between the point the view from which the image represents and the point in the scene represented by that pixel of the image, means for computing an estimate of the contribution of backscattered light to the brightness of each pixel of the image from the model, means for subtracting the estimated contribution for each pixel from the brightness of that pixel of the image to produce a modified brightness for that pixel, and means for reconstructing the image to form the enhanced image with each pixel of the enhanced image having a brightness which is a function of the modified brightness of that pixel.
• the present invention is based on the realisation that the mean backscattered flux component in an image of a scene obscured for example by cloud will vary according to the distance between the point from which the image was captured and the points in the terrain represented in the image. Depending on the type of view, this distance (or depth) will vary across the image. This is particularly the case with aircraft flying generally horizontally across the surface of the earth.
• the invention considers these image-plane variations of depth and as a result is robust to variations in mean backscatter.
• the model may be defined by reference to a plot of the depth/brightness relationship for the individual image to be enhanced.
• the brightness of each pixel ofthe enhanced image may be computed from the modified brightness of that pixel scaled to restore the image contrast, for example by computing the contribution of light from the scene to the brightness of each pixel of the image on the basis of the model, dividing the modified brightness for each pixel by the estimated contribution from the scene for that element, and multiplying the resultant by the a constant to determine the brightness of the pixel in the enhanced image.
• Figure 1 is a single unprocessed optical image captured by an airborne camera
• Figure 2 is a histogram of the image of Figure 1 plotting the intensity against the number of pixels in the image having particular intensities;
• Figure 3 is an image of the same scene as that represented in Figure 1 but representing the image resulting from the motion-compensated average of ten frames captured sequentially:
• Figure 4 is a histogram corresponding to that of Figure 2 but relating to the image of Figure 3;
• Figure 5 plots the depth to brightness relationship of the pixels of the image of Figure 1;
• Figure 6 plots the depth to brightness relationship of the pixels of Figure 3;
• Figure 7 is an image derived from the image of Figure 3 in accordance with the present invention:
• Figure 8 is a further unprocessed image captured by an airborne camera
• Figures 9 and 10 are processed images derived from the image of Figure 8 using respectively the method of the present invention and a conventional image processing method;
• Figure 11 is a further unprocessed image captured by an airborne camera.
• Figure 12 is an image derived from a series of images including that of Figure 1 1 in accordance with the invention.
• the method of image enhancement described below has three steps, that is image averaging, parameter estimation, and contrast transformation.
• Image averaging techniques used are conventional.
• Figure 1 represents a single unprocessed image generated by the camera.
• the image is of a scene including a major road, ⁇ ehicles travelling on that road, bridges across the road, and various terrain features to both sides of the road. The features ofthe scene are obscured by low level cloud.
• a series of ten images was taken by the camera as the aircraft travelled relative to the imaged scene.
• Motion- compensated averaging was then applied, the averaging being computed over the ten image frames.
• the image averaging is performed such that the "averaged" image at frame N is derived from the sum of a number of previous frames, using a geometric transformation to correct for the camera movements.
• a time-averaged image is maintained which always reflects the current camera position and orientation.
• Each pixel in this averaged image corresponds to some particular point of the terrain which contributes to the image. Assuming worldspace (terrain) co-ordinates (x, y, z) this can be written as the sum:
• the sensor noise is reduced by a factor of approximately
• Figure 2 is a histogram based on the image of Figure 1 and representing the number of pixels in that image having the identified intensities. This histogram is dominated by the backscattered light from the cloud.
• Figure 3 shows the image which results from the motion-compensated average of ten frames
• Figure 4 is a histogram corresponding to that of Figure 2 but in respect of the image of Figure 3.
• Figure 4 shows a narrower distribution of values around the mean gray level than Figure 2.
• the light flux which has been reflected from the terrain is contained within this relatively narrow peak and can now be recovered by contrast enhancement.
• contrast enhancement algorithms such as variance normalisation or histogram equalisation are suitable for this purpose.
• the backscatter contribution is estimated in each pixel in the image. This is achieved by considering the depth to brightness relationship of many of the pixels in the averaged image. A parametric model is then fitted to the depth/brightness data using a numerical data-fitting algorithm. The resulting model parameters are then used to calculate backscatter at any image pixel. This method of backscatter estimation also provides parameters for contrast enhancement.
• Figure 5 plots the depth to brightness relationship of the pixels of the image of Figure 1.
• Figure 6 plots the same data for Figure 3.
• the motion-compensated average image of Figure 3 shows less scatter and therefore represents a better starting point for the application of the present invention.
• t(d) T exp(-Kd) (3) where T is a constant depending on C and the nature of the terrain scene.
• the model fit is carried out by using a numerical optimisation algorithm to determine values for CM . C ⁇ . and K such that the total least squares difference, defined by
• the estimate for t(ij) takes account of the attenuation of the terrain-reflected light by the scattering medium.
• Figures 8 - 10 illustrate the superiority of the image enhancement provided by the backscatter estimation and image enhancement algorithms over a conventional contrast stretch algorithm (histogram equalisation) in fairly good visibility conditions.
• Figure 8 shows an original image of an airstrip with two runways, with little attenuation of the image by mist.
• Figure 9 shows the image processed using the above described algorithms, and Figure 10 shows the image after processing by the contrast stretch algorithm.
• the contrast stretched image of Figure 10 has an artificial appearance, whereas Figure 9 has a natural appearance.
• the solid angle ⁇ k is a fixed geometric constant associated with the position ofthe sensor element in the image plane. ⁇ k may be calculated from
• A is the active area of a single sensor element, /is the focal length, and x k and y k are the x and y offsets of pixel k from the optical centre in the image plane. Compensation for variation ofthe solid angle across the image is achieved by dividing each pixel value by the solid angle.
• Figure 11 shows one of a series of three images captured with an airborne camera
• Figure 12 shows an enhanced image obtained by reversing the gamma- encoding by the camera to make the pixel values proportional to the brightness of the terrain, averaging the three images, and then compensating for the variation of solid angle.
• a similar enhanced image to that depicted in Figure 12 was achieved by processing each ofthe three images individually and averaging the enhanced images.
• a problem with the averaging process whereby several images are summed together to improve the ratio of signal-to-noise is that uncertainties in the position of the aircraft give rise to blurring in the foreground of the averaged image.
• the amount of averaging may be varied with respect to the depth ofthe image. In such an arrangement, pixels relating to very distant terrain points (near the horizon) are averaged many times, whilst pixels relating to points close to the camera (i.e. the bottom of the image) may be averaged over only two or three images, or may be bypassed by the averaging algorithm.
• This method is particularly effective because the signal-to-noise ratio (SNR) in the enhanced image is a strong function of depth and tends to be much lower for very distant points, whereas the SNR at lower depths is often much greater.
• SNR signal-to-noise ratio
• y[n] o ⁇ [n] + ( ⁇ - a )y[n - ⁇ ], (11)
• x[n] is the wth image of the sequence
• v[ «] is the filtered version of x[n]
• v[ «-l] is the filtered version of the ( «-l)th image.
• the noise reduction properties of the filter (11) depend on the choice of the constant ⁇ ; the smaller the value for ⁇ , the
• the SNR may be estimated in various ways.
• One available method is based on the assumption that the noise in the backscatter is multiplicative Gaussian and that the noise in the image is the sum of the backscatter noise and noise generated in the sensor itself.
• the SNR as a function of depth is then given by
• an image averaging routine is performed prior to estimation of the backscatter contribution to the image and contrast enhancement
• the backscatter estimation and contrast enhancement routines may be applied to single images from a series of which an average image is subsequently generated.
• One advantage of performing the processing in this order is the avoidance of the problem of image degradation caused by depth values corresponding to particular pixels varying significantly over the averaging interval.
• backscatter estimation and image enhancement algorithms may be applied to single images, with no image averaging routine being used either before or after the enhancement.
• the backscattered contribution could be estimated by applying a low-pass filter to the averaged image.
• this estimate would be degraded by any low spatial frequency components in the terrain signals.
• the size of the filter kernel would have to be small with respect to the expected variation in depth. This would mean that the backscatter estimates from the filter would be subject to a greater degree of random uncertainty that with the estimation method described above.
• contrast transformation that is the final generation of the enhanced image from the estimated backscatter and transmitted terrain contributions.
• the terrain contribution t(i,j) calculated from equation (8) will yield information as to the reflectivity of the elements included in the image.
• the reflectance factors corresponding to different terrain cover, for example grass, trees and water, are fixed.
• the terrain contribution t(i,j) may be scaled so that it gives a direct estimate for the local reflectance factor for pixel (i,j).
• a simple computer program could store the calculated values of t(i,j) and then convert them into possible terrain properties.
• the invention is applicable to infra-red images as well as to visible light.

Landscapes

• Engineering & Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Multimedia (AREA)
• Remote Sensing (AREA)
• Astronomy & Astrophysics (AREA)
• Image Processing (AREA)
• Photoreceptors In Electrophotography (AREA)
• Apparatus For Radiation Diagnosis (AREA)
• Glass Compositions (AREA)
• Lubrication Of Internal Combustion Engines (AREA)
• Magnetic Resonance Imaging Apparatus (AREA)
• Transition And Organic Metals Composition Catalysts For Addition Polymerization (AREA)

Priority Applications (7)

Applications Claiming Priority (2)

Publications (1)

Family

ID=10777955

Family Applications (1)

Country Status (12)

Cited By (5)

Families Citing this family (36)

Citations (4)

Family Cites Families (6)

• 1995
  • 1995-07-19 GB GB9514834A patent/GB2303511A/en not_active Withdrawn
• 1996
  • 1996-07-12 EP EP96924053A patent/EP0839361B1/en not_active Expired - Lifetime
  • 1996-07-12 AU AU64638/96A patent/AU6463896A/en not_active Abandoned
  • 1996-07-12 CA CA002227321A patent/CA2227321A1/en not_active Abandoned
  • 1996-07-12 US US08/981,959 patent/US6462768B1/en not_active Expired - Fee Related
  • 1996-07-12 DK DK96924053T patent/DK0839361T3/da active
  • 1996-07-12 JP JP50639997A patent/JP3898224B2/ja not_active Expired - Fee Related
  • 1996-07-12 DE DE69604670T patent/DE69604670T2/de not_active Expired - Lifetime
  • 1996-07-12 ES ES96924053T patent/ES2140110T3/es not_active Expired - Lifetime
  • 1996-07-12 WO PCT/GB1996/001657 patent/WO1997004417A1/en not_active Ceased
  • 1996-07-12 AT AT96924053T patent/ATE185637T1/de not_active IP Right Cessation
• 1999
  • 1999-12-17 GR GR990403265T patent/GR3032179T3/el unknown

Patent Citations (4)

Non-Patent Citations (3)

Also Published As

Similar Documents

Legal Events