Image processing apparatus

The present invention relates to the field of time-lapse photography.

In time-lapse photography, a camera is periodically operated to photograph a subject, and then the recorded images are viewed, usually in the form of a “movie”, over a period of time which is much shorter than the period over which the photographs were taken. This enables a slowly occurring change in the subject (which would normally occur at a speed too slow to see) to be studied at a faster than normal rate.

Considerable work has been carried out in the field of time-lapse photography to improve the functionality of time-lapse photographic equipment to give the user more control over the recording process, for example by controlling the exposure duration and exposure interval etc (see for example U.S. Pat. No. 3,790,262, U.S. Pat. No. 4,466,721 and U.S. Pat. No. 4,052,126). However, existing systems still suffer from many problems, in particular inflexibility, difficulty of use and cost.

The present invention has been made with the above problems in mind.

According to the present invention, there is provided a method or apparatus for generating image data for a time-lapse sequence of a scene in which image data recorded of the scene at different times is processed in such a way that the images need not have been recorded at the same position and with the same viewing direction.

This enables the photographer to revisit this scene at the required times without having to leave the camera in place in between, and, if required, to exercise full manual control of the camera to record the images.

The present invention also provides an image processing apparatus or method for processing input images of a scene recorded at different times and with different viewing characteristics in which processing is carried out to correct for the different viewing characteristics.

The present invention further provides an image processing apparatus or method in which image data defining images recorded of the scene at different times is processed to align the input images and to generate data for a time-lapse sequence at times between the times at which the input images were recorded by calculating data defining synthetic images.

The present invention further provides instructions, both in recorded and signal form, for configuring a programmable processing apparatus to perform such a method or to become configured as such an apparatus.

Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings in which:

FIG. 1 is a block diagram showing an example of notional functional components within a processing apparatus of an embodiment of the invention;

FIGS. 2 a , 2 b , 2 c and 2 d show examples of input images of a scene recorded at different times to be processed by the apparatus shown in FIG. 1 to generate a time-lapse sequence of images;

FIG. 3 shows the processing operations performed by the apparatus in FIG. 1 ;

FIG. 4 shows the processing operations performed at step S 30 in FIG. 3 to calculate transformations to register the input images;

FIG. 5 shows the processing operations performed at step S 40 in FIG. 3 to generate interpolated image data; and

FIG. 6 schematically shows the effect of registering a pair of input images at step S 200 in FIG. 5 .

Referring to FIG. 1 , an embodiment of the invention comprises a processing apparatus 2 , such as a personal computer, user input devices 4 , such as a keyboard, mouse etc., and a display device 6 , such as a conventional personal computer monitor.

The processing apparatus 2 is programmed to operate in accordance with programming instructions input for example as data stored on a data storage medium, such as disk 8 , and/or as a signal 10 input to the processing apparatus, for example from a remote database, over a datalink (not shown) such as the Internet, and/or entered by a user via a user input device 4 .

The programming instructions comprise instructions to cause the processing apparatus 2 to become configured to process images of a scene taken at different times and from different viewing positions and/or viewing directions and to generate from this input image data image data defining a sequence of images for display which represents an evolving representation of the part of the scene which is present in all of the input images between the times at which the input images were recorded (that is, a so-called “time-lapse” sequence).

When programmed by the programming instructions, processing apparatus 2 effectively becomes configured into a number of functional units for performing processing operations. Examples of such functional units and their interconnections are shown in FIG. 1 . The illustrated units and interconnections in FIG. 1 are, however, notional and are shown for illustration purposes only to assist understanding; they do not necessarily represent the exact units and connections into which the processor, memory etc. of the processing apparatus become configured.

Referring to the functional units shown in FIG. 1 , central controller 20 processes inputs from the user input devices 4 , and also provides control and processing for a number of the other functional units. Memory 22 is provided for use by central controller 20 and other functional units.

Image data store 28 stores the image data representing pictures of a scene input to the processing apparatus 2 . Image data registration processor 30 and interpolator 32 perform processing of the image data stored in image data store 28 to generate image data defining images to be displayed in a “time-lapse” sequence of images. Image data generated by interpolator 32 is stored in the interpolated image data store 36 .

Image renderer 40 reads image data from the image data store 28 and the interpolated image data store 36 and generates pixel data for images to be displayed. The pixel data generated by image renderer 40 is written to frame buffer 42 for display on the display device 6 at a video rate.

Output processor 44 generates data for output to another processing apparatus conveying sufficient information that the time lapse sequence of images produced by processing apparatus 2 can be generated from the data by the receiving apparatus.

FIGS. 2 a , 2 b , 2 c and 2 d show examples of input images to be processed by processing apparatus 2 in this embodiment.

Each of the images shown in FIGS. 2 a , 2 b , 2 c and 2 d comprises a picture of a scene (in this example, a garden) taken at different times. More particularly, the image shown in FIG. 2 a was recorded at the earliest time of the four images, the image shown in FIG. 2 b was recorded next, followed by the image shown in FIG. 2 c and finally the image shown in FIG. 2 d . As can be seen from FIGS. 2 a , 2 b , 2 c and 2 d , changes have occurred in the scene between the times at which the images were recorded (in this example, plants, shrubs and weeds have grown in the garden). Processing apparatus 2 is arranged to process the image data defining these input images to generate many more images representing the scene at times between those at which the input images were recorded. The input images and the generated images are then used to display a sequence of images to a user at video rate (or at a user-controlled speed, e.g. fast forward, reverse, pause etc.) showing how the scene changes over time (that is, in this example, how the plants, shrubs and weeds grow).

In this embodiment, it is not necessary for the input images to be recorded at equally spaced time intervals that is, the time between the recording of the image shown in FIG. 2 a and the image shown in FIG. 2 b can be different from the time between the time at which the image shown in FIG. 2 b was recorded and the time at which the image shown in FIG. 2 c was recorded. In addition, the number of input images is not limited to four, and any number can be input, provided there are at least two.

In addition, as will be seen from FIGS. 2 a , 2 b , 2 c and 2 d , the input images, while showing substantially the same scene, are not necessarily recorded from the same viewing position or viewing direction. For example, referring to FIGS. 2 a and 2 b , and in particular the paved patio area 50 , it will be seen that the image in FIG. 2 b was recorded at a viewing position further away from the fence 52 than the viewing position at which the image in FIG. 2 a was recorded, and with a viewing direction to the right of the viewing direction with which the image in FIG. 2 a was recorded.

It will be appreciated that, while the input images shown in the example of FIGS. 2 a , 2 b , 2 c and 2 d were recorded over a period of several months, the input images could be recorded over any timescale.

Regarding the number of input images, and the time between the times at which they were recorded, the additional images generated by processing apparatus 2 will be more accurate (that is, will better resemble the true state of the scene at the time the generated image is supposed to represent) the closer the times at which the input images were recorded so that the changes in the scene between the input images are smaller. Similarly, the accuracy of the images generated by processing apparatus 2 will be higher if the input images are recorded under lighting conditions which are approximately the same when each image is recorded.

FIG. 3 shows the processing operations performed in this embodiment by processing apparatus 2 .

Referring to FIG. 3 , at step S 110 , the image data input to processing apparatus 2 defining the input images is stored in the image data store 28 .

At step S 20 , central controller 20 causes a message to be displayed on display device 6 requesting the user to enter data via the user input devices 4 defining the order in which the input images were recorded, the approximate time between the times at which the input images were recorded, and the required time for which the generated time-lapse sequence is to run when displayed to the user. Information input by the user in response to this request is stored in memory 22 at step S 25 .

At step S 30 , central controller 20 and image data registration processor 30 perform processing to calculate transformations suitable to register the input images.

FIG. 4 shows the processing operations performed by central controller 20 and image data registration processor 30 in this embodiment to calculate the transformations at step S 30 .

Referring to FIG. 4 , at step S 100 , central controller 20 uses the information stored at step S 25 defining the order of the input images to read the image data from image data store 28 defining the next pair of consecutive input images (this being the first and second input images in time of recording order the first time step S 100 is performed) and displays the pair of input images on display device 6 via image renderer 40 and frame buffer 42 .

At step S 110 , central controller 20 causes a message to be displayed on display device 6 asking the user to identify four pairs of matching points (that is, the point in each displayed image which corresponds to the same real-world point) on each plane present in the displayed images, each point being a stationary point (that is, a point on an object which is not moving, an example of a moving object being a growing plant etc.). Preferably, the user is instructed to select the four points on each plane so that the points are spread through the area of the plane which is visible in both of the displayed images. The user then inputs this information by using an input device such as a mouse to identify a point in the first displayed image and the corresponding (matching) point in the second displayed image. Referring to FIGS. 2 a and 2 b by way of example, the patio 50 and lawn 54 lie in one plane, while the fence 52 lies in a second plane. Accordingly, at step S 110 , the user identifies four pairs of matching points on the patio 50 and/or lawn 54 (such as points 56 a and 56 b , 58 a and 58 b , 60 a and 60 b , and 62 a and 62 b ) and four pairs of matching points on fence 52 (such as points 58 a and 58 b , 60 a and 60 b , 64 a and 64 b , and 66 a and 66 b ).

At step S 120 , the coordinates of the matching points identified by the user at step S 110 are stored in memory 22 .

At step S 130 , central controller 20 causes a message to be displayed on display device 6 requesting the user to identify the boundary of each plane in the images on which matching points were identified at step S 110 . In this embodiment, this is performed by the user moving a cursor around the boundary of each plane using a user input device 4 . Referring to FIG. 2 by way of example, the user defines the extent of the plane containing the patio 50 and lawn 54 by moving a cursor along the image edges 80 , 82 and 84 and along the boundary 86 between the lawn 54 and the fence 52 . To define the plane of the fence 52 , the user moves the cursor along the image edges 88 , 90 and 92 and along the boundary 86 .

At step S 140 , central controller 20 stores the coordinates of the boundaries in each image input by the user at step S 130 .

At step S 150 , image data registration processor 30 reads the coordinates of the matching points stored at step S 120 and calculates the respective transformation for each plane which will register the four pairs of matching points in the images, that is, the transformation which will transform the four points in the second image of the pair so that they align with the four points in the first image of the pair. In this embodiment, image data registration processor 30 calculates each transformation by calculating the 3×3 matrix H which satisfies the following equation:

H ⁡ ( x i y i 1 ) = ( u i v i w i ) ( 1 )

where: (x i , y i ) are the coordinates of the points in one input image and (x′ i ,y′ i ) are the coordinates of the corresponding (matched) points in the other input image (i=1, 2, 3, 4); and

x′ i =u i /w i and y′ i =v i /w i .

Equation (1) can be written as:

( h 0 ⁢ h 1 ⁢ h 2 h 3 ⁢ h 4 ⁢ h 5 h 6 ⁢ h 7 ⁢ h 8 ) ⁢   ⁢ ( x i y i 1 ) = ( x i ′ ⁢ w i y i ′ ⁢ w i w i ) ( 2 )

This gives:

h 0 x i +h 1 y i +h 2 =x′ i w i   (3)

h 3 x i +h 4 y i +h 5 =y′ i w i   (4)

h 6 x i +h 7 y i +h 8 =w i   (5)

However, wi can be eliminated to get 2 equations per point, as follows:

h 0 x i +h 1 y i +h 2 =x′ i ( h 6 x i +h 7 y i +h 8 )  (6)

h 3 x i +h 4 y i +h 5 =y′ i ( h 6 x i +h 7 y i +h 8 )  (7)

Rearranging gives:

h 0 x i +h 1 y i +h 2 −h 6 x i x′ i −h 7 y i x′ i −h 8 x′ i =0  (8)

h 3 x i +h 4 y i +h 5 −h 6 x i y′ i −h 7 y i y′ i −h 8 y′ i =0  (9)

Putting the four sets of two equations together gives:

( x 1 y 1 1 0 0 0 - x 1 ⁢ x 1 ′ - y 1 ⁢ x 1 ′ - x 1 ′ 0 0 0 x 1 y 1 1 - x 1 ⁢ y 1 ′ - y 1 ⁢ y 1 ′ - y 1 ′ x 2 y 2 1 0 0 0 - x 2 ⁢ x 2 ′ - y 2 ⁢ x 1 ′ - x 2 ′ 0 0 0 x 2 y 2 1 - x 2 ⁢ y 2 ′ - y 2 ⁢ y 2 ′ - y 2 ′ x 3 y 3 1 0 0 0 - x 3 ⁢ x 3 ′ - y 3 ⁢ x 3 ′ - x 3 ′ 0 0 0 x 3 y 3 1 - x 3 ⁢ y 3 ′ - y 3 ⁢ y 3 ′ - y 3 ′ x 4 y 4 1 0 0 0 - x 4 ⁢ x 4 ′ - y 4 ⁢ x 4 ′ - x 4 ′ 0 0 0 x 4 y 4 1 - x 4 ⁢ Y 4 ′ - y 4 ⁢ y 4 ′ - y 4 ′ ) ⁢   ⁢ ( h 0 h 1 h 2 h 3 h 4 h 5 h 6 h 7 h 8 ) = 0

The matrix M on the left is an 8×9 matrix but can be made square (9×9) by multiplying both sides by the transpose of M, as follows (note that, if there are more than four points, more rows can be added to M and M T M will still be a 9×9 matrix):

( M T M ) h =0  (11)

Equation (11) is in the well known Eigensystem form:

Ax=λx   (12)

with A=M T M (a symmetric 9×9 matrix) and λ=0. This cannot be solved analytically, but can be solved numerically in a number of ways. For example, in this embodiment, image data registration processor 30 and central controller 20 are arranged to perform processing to solve the equation using the method described in Chapter 11 of “Numerical Recipes in C” ISBN 0-521-43108-5. More particularly, image data registration processor 30 and central controller 20 are arranged to use the routines “jacobi” and “eigsrt” in the above-referenced “Numerical Recipes in C” to calculate the eigenvectors (x's) of A (in general there will be 9 distinct solutions) and to take the solution with the eigenvalue (x) closest to zero. This vector then contains the 9 elements of the matrix H that defines the required image transformation.

Referring again to FIG. 4 , at step S 160 , central controller 20 determines whether there is data stored in image data store 28 defining a further input image, and steps S 100 to S 150 are repeated until each pair of consecutive input images has been processed in the manner described above. More particularly, after processing the first and second input images, the second and third input images are processed in the same way, followed by the third and fourth input images etc.

Referring again to FIG. 3 , at step S 40 , image data registration processor 30 , interpolator 32 and central controller 20 perform processing to generate data defining images representing the state of the scene between the times at which the input images were recorded.

FIG. 5 shows the processing operations performed at step S 40 .

Referring to FIG. 5 , at step S 200 , image data registration processor 30 applies the transformations calculated at step S 150 to the input image data so as to register the next pair of input images (this being the first pair the first time step S 200 is performed).

FIG. 6 shows the effect of registering the input image shown in FIG. 2 b with the input image shown in FIG. 2 a.

Referring to FIG. 5 , the effect of registering the input images is to overlap the input image 100 representing the input image shown in FIG. 2 b with the input image 102 representing the input image shown in FIG. 2 a such that the matching points identified by the user at step S 110 are aligned. In FIG. 5 , the parts of image 100 shaded with dots represent the parts of the image shown in FIG. 2 b which are not present in the image shown in FIG. 2 a Similarly, the parts of image 102 shaded with diagonal lines represent the parts of the image shown in FIG. 2 a which are not present in the image shown in FIG. 2 b . The unshaded portion 104 represents the area which is present in both the input image shown in FIG. 2 a and the input image shown in FIG. 2 b.

Referring again to FIG. 5 , at step S 210 , interpolator 32 and central controller 20 use the information stored in memory 22 at step S 25 defining the approximate time between the times at which the input images were recorded and the required run time for the time-lapse sequence of images to be generated, to determine the number of intervening images in the time-lapse sequence between the pair of input images currently being considered, for which input image data is to be generated.

At step S 220 , interpolator 32 and central controller 20 generate image data defining each intervening image in the time-lapse sequence between the pair of input images currently being considered. More particularly, in this embodiment, for each pixel in the region where the input images overlap after registration (that is, region 104 in FIG. 6 ) interpolator 32 and central controller 20 perform interpolation between the value of the pixel in the first input image of the pair and the value of the pixel in the second input image of the pair. For each given pixel in the region 104 , this interpolation is performed to generate a value for each intervening image in the time-lapse sequence in dependence upon the position of the intervening image in the sequence. In this embodiment, interpolator 32 and central controller 20 are arranged to calculate the pixel values using linear interpolation. The pixel values generated at step S 220 are stored in the interpolated image data store 36 .

At step S 230 , central controller 20 determines whether there is another input image to be processed, and steps S 200 to S 230 are repeated until each successive pair of input images have been processed in the manner described above. When these steps have been performed for all of the input images, image data has been generated and stored for images defining a time-lapse sequence.

Referring again to FIG. 3 , at step S 50 , central controller 20 and image renderer 40 use the image data defining the input images stored in the image data store 28 and the image data defining the interpolated images stored in the interpolated image data store 36 to display the time-lapse sequence to the user on display device 6 .

In this embodiment, image renderer 40 and central controller 20 are arranged to perform processing such that, when an input image is to be displayed as part of the time-lapse sequence, only the pixel values for the region 104 shown in FIG. 6 in which the registered images overlap are read from image data store 28 and displayed on monitor 6 . This prevents a change in image size when an input image is displayed followed by interpolated images.

In addition, or instead, output processor 44 outputs signals defining information from which the time-lapse sequence can be generated and displayed by a processing apparatus which receives the output signals. In this embodiment, output processor 44 is operable to output signals defining the image data of the input images stored in image data store 28 and the interpolated images stored in the interpolated image data store 36 . In addition, output processor 44 is operable to output image data from image data store 28 defining the input images, information defining the transformations calculated by image data registration processor 30 to register the input images, and information defining the type of interpolation performed by interpolator 32 . In this way, a receiving apparatus which itself has an interpolator can reconstruct the image data for the time-lapse sequence, and it is unnecessary for output processor 44 to output pixel data for each interpolated image.

A number of modifications are possible to the embodiment described above.

For example, in the embodiment above, image data registration processor 30 is arranged to calculate transformations to register the input images using matching features in the images identified by the user. However, image data registration processor 30 may itself be arranged to identify matching features in the input images using conventional image processing techniques, such as the feature matching techniques described in “Adaptive Least Squares Correlation: A Powerful Image Matching Technique” A. W. Gruen and S. Afr in Journal of Photogrammetry, Remote Sensing and Cartograhy 14(3)1985 pages 175-187 and/or the entropy matching techniques described in “Alignment by Maximisation of Mutual Information”, P. Viola and W. Wells in ICCV 5, Jun. 20-23 1995, MIT, ISBN 0-8186-7042-8 pages 16-23. As will be appreciated by the skilled person, for some of these techniques, the closer the input images are taken to the same viewing position and same viewing direction, the easier it will be to perform matching and the more accurate the matches will be. In addition, image data registration processor 30 may be arranged to calculate the transformations on the basis of a combination of user identified matches and matches calculated itself using image processing techniques.

In the embodiment above, in steps S 100 -S 160 (FIG. 4 ), each input image is registered to the preceding input image using points matched in the pair of images by the user. Instead, all of the input images may be displayed to the user simultaneously and the user requested to match points which are present in all of the input images. In this way, each input image would then be registered to the first input image, rather than the preceding input image. Of course, this effect is also achieved if the user matches the same points between each pair of images (as illustrated in FIGS. 2 a , 2 b , 2 c and 2 d ).

In the embodiment above, each plane in the input images is registered, which increases the accuracy of the interpolated image data. However, it is possible to register the images using matching points which lie on just one plane, or indeed matching points which do not necessarily lie on the same plane, with a subsequent loss of accuracy in the interpolated data.

In the embodiment above, each plane in the input images is registered using four pairs of matching points. By using four pairs of points, a full projective transformation can be calculated as described in the embodiment above, which will take account of all perspective effects in the images. However, it is possible to perform registration using fewer pairs of matching points. For example, using three pairs of matching points, an affine transformation could be calculated, in which case perspective effects in the images would be lost.

In the embodiment above, at step S 200 (FIG. 5 ), an input image is transformed to register it with the preceding image, and subsequently at step S 220 , interpolation is performed for each pixel in the overlapping region of the resulting registered input images. However, as will be appreciated by the skilled person, it is possible to perform these steps together so as to transform and generate interpolated values for a pixel using a single mathematical operation.

In the embodiment above, at step S 40 the image data for the intervening images in the time-lapse sequence is calculated for each pixel in each intervening image by linear interpolation between the value of the pixel in one input image and the value of the pixel in the following input image. This linear interpolation assumes that each input image will be displayed as one frame of the time-lapse video sequence. However, to ensure that the input images are more prominent in the time-lapse sequence, it is possible to repeatedly display each input image over a number of consecutive frames in the time-lapse sequence. In this case, the number of intervening images to be calculated by interpolation is reduced and the linear interpolation parameters are adjusted accordingly.

In the embodiment above, at step S 40 , the image data for intervening images in the time-lapse sequence is generated by linear interpolation. However, other types of interpolation could be performed.

In the embodiment above, output processor 44 is operable to output the pixel data of the input images together with the registration transformations and information defining the type of interpolation performed by interpolator 32 . However, instead, output processor 44 may be arranged to output pixel data for a number of interpolated images together with information defining the interpolation so that further images can be generated by the receiving processor.

The apparatus in the embodiment above may be arranged to perform processing to update the time-lapse sequence if the user adds further input images, either to improve the accuracy of an existing sequence, or to extend an existing sequence using input images recorded at later times than those originally processed.

In the embodiment above, processing is performed by a computer using processing routines defined by programming instructions. However, some, or all, of the processing could be performed using hardware.

Other changes and modifications can be made without departing from the spirit and scope of the invention.