scispace - formally typeset
Search or ask a question
Journal ArticleDOI

Destriping multisensor imagery with moment matching

01 Jan 2000-International Journal of Remote Sensing (Taylor & Francis Group)-Vol. 21, Iss: 12, pp 2505-2511
TL;DR: An alternative algorithm is suggested which matches the gain and offset of each sensor to typical values, and which is resistant to the effects of outliers.
Abstract: Image destriping is necessary due to sensor-to-sensor variation within instruments. This has most often been done by assuming that each sensor views a statistically similar subimage, and a histogram of each sensor's response is made to match the overall histogram. Histogram matching shows sensitivity to violations of the similarity assumption. An alternative algorithm is suggested which matches the gain and offset of each sensor to typical values, and which is resistant to the effects of outliers. Tests on a sample image show the moment matching algorithm reduces the variance between sensors to a greater degree than histogram matching.
Citations
More filters
Journal ArticleDOI
TL;DR: A novel graph-regularized low-rank representation (LRR) destriping algorithm is proposed by incorporating the LRR technique and can both remove striping noise and achieve cleaner and higher contrast reconstructed results.
Abstract: Hyperspectral image destriping is a challenging and promising theme in remote sensing. Striping noise is a ubiquitous phenomenon in hyperspectral imagery, which may severely degrade the visual quality. A variety of methods have been proposed to effectively alleviate the effects of the striping noise. However, most of them fail to take full advantage of the high spectral correlation between the observation subimages in distinct bands and consider the local manifold structure of the hyperspectral data space. In order to remedy this drawback, in this paper, a novel graph-regularized low-rank representation (LRR) destriping algorithm is proposed by incorporating the LRR technique. To obtain desired destriping performance, two sides of performing destriping are included: 1) To exploit the high spectral correlation between the observation subimages in distinct bands, the technique of LRR is first utilized for destriping, and 2) to preserve the intrinsic local structure of the original hyperspectral data, the graph regularizer is incorporated in the objective function. The experimental results and quantitative analysis demonstrate that the proposed method can both remove striping noise and achieve cleaner and higher contrast reconstructed results.

287 citations


Cites background or methods from "Destriping multisensor imagery with..."

  • ...In our experiment, we compare the destriping results of the proposed method with the moment-matching method [9], the total variation (TV)-based denoising method [39], and the K-singular value decomposition (Ksvd) method [40] on the four test subimages....

    [...]

  • ...Moment-matching-based methods [9], [18] often assume that this relation can be described by a linear function....

    [...]

  • ...Therein, moment matching [9], [18] and histogram matching [10], [19]– [21] are typical examples....

    [...]

  • ...For the sake of comparison, the moment-matching method is implemented according to [9]....

    [...]

  • ...2–5 show the destriping results of the four subimages with the moment-matching method [9], the TV-based denoising method [39], the Ksvd method [40], and the proposed method, respectively....

    [...]

Journal ArticleDOI
TL;DR: Wu et al. as discussed by the authors employed multi-temporal and multi-sensor fusion methods for a long-term and fine-scale summer SUHI analysis of the city of Wuhan in China.

254 citations

Journal ArticleDOI
TL;DR: The proposed algorithm has been tested using moderate resolution imaging spectrometer images for destriping and China-Brazil Earth Resource Satellite and QuickBird images for simulated inpainting and the results and quantitative analyses verify the efficacy of this algorithm.
Abstract: Remotely sensed images often suffer from the common problems of stripe noise and random dead pixels. The techniques to recover a good image from the contaminated one are called image destriping (for stripes) and image inpainting (for dead pixels). This paper presents a maximum a posteriori (MAP)-based algorithm for both destriping and inpainting problems. The main advantage of this algorithm is that it can constrain the solution space according to a priori knowledge during the destriping and inpainting processes. In the MAP framework, the likelihood probability density function (PDF) is constructed based on a linear image observation model, and a robust Huber-Markov model is used as the prior PDF. The gradient descent optimization method is employed to produce the desired image. The proposed algorithm has been tested using moderate resolution imaging spectrometer images for destriping and China-Brazil Earth Resource Satellite and QuickBird images for simulated inpainting. The experiment results and quantitative analyses verify the efficacy of this algorithm.

240 citations


Cites background or methods from "Destriping multisensor imagery with..."

  • ...[9] use a moment matching method to remove the stripes in Landsat thematic mapper (TM) remote sensing images....

    [...]

  • ...In this paper, the moment matching method is used [9]....

    [...]

  • ...In this paper, it is assumed that the degradation process can be linearly described as in [9] and [17], but the existence of linear-assumption error is permitted....

    [...]

  • ...Another destriping approach examines the distribution of digital numbers (DNs) for each sensor and adjusts this distribution to some reference distribution [7], [9]....

    [...]

Journal ArticleDOI
TL;DR: An overview of the techniques developed in the past decade for hyperspectral image noise reduction is provided, and the performance of these techniques by applying them as a preprocessing step to improve a hyperspectrals image analysis task, i.e., classification.
Abstract: Hyperspectral remote sensing is based on measuring the scattered and reflected electromagnetic signals from the Earth’s surface emitted by the Sun. The received radiance at the sensor is usually degraded by atmospheric effects and instrumental (sensor) noises which include thermal (Johnson) noise, quantization noise, and shot (photon) noise. Noise reduction is often considered as a preprocessing step for hyperspectral imagery. In the past decade, hyperspectral noise reduction techniques have evolved substantially from two dimensional bandwise techniques to three dimensional ones, and varieties of low-rank methods have been forwarded to improve the signal to noise ratio of the observed data. Despite all the developments and advances, there is a lack of a comprehensive overview of these techniques and their impact on hyperspectral imagery applications. In this paper, we address the following two main issues; (1) Providing an overview of the techniques developed in the past decade for hyperspectral image noise reduction; (2) Discussing the performance of these techniques by applying them as a preprocessing step to improve a hyperspectral image analysis task, i.e., classification. Additionally, this paper discusses about the hyperspectral image modeling and denoising challenges. Furthermore, different noise types that exist in hyperspectral images have been described. The denoising experiments have confirmed the advantages of the use of low-rank denoising techniques compared to the other denoising techniques in terms of signal to noise ratio and spectral angle distance. In the classification experiments, classification accuracies have improved when denoising techniques have been applied as a preprocessing step.

208 citations


Cites methods from "Destriping multisensor imagery with..."

  • ...A moment matching (MM) technique [56,57] has been adapted for HSI striping noise removal in [25] by exploiting the spectral correlations....

    [...]

Journal ArticleDOI
TL;DR: This paper tentatively categorizes the stripes in remote sensing images in a more comprehensive manner and proposes to treat the multispectral images as a spectral-spatial volume and pose an anisotropic spectral- spatial total variation regularization to enhance the smoothness of solution along both the spectral and spatial dimension.
Abstract: Multispectral remote sensing images often suffer from the common problem of stripe noise, which greatly degrades the imaging quality and limits the precision of the subsequent processing. The conventional destriping approaches usually remove stripe noise band by band, and show their limitations on different types of stripe noise. In this paper, we tentatively categorize the stripes in remote sensing images in a more comprehensive manner. We propose to treat the multispectral images as a spectral-spatial volume and pose an anisotropic spectral-spatial total variation regularization to enhance the smoothness of solution along both the spectral and spatial dimension. As a result, a more comprehensive stripes and random noise are perfectly removed, while the edges and detail information are well preserved. In addition, the split Bregman iteration method is employed to solve the resulting minimization problem, which highly reduces the computational load. We extensively validate our method under various stripe categories and show comparison with other approaches with respect to result quality, running time, and quantitative assessments.

187 citations


Cites background from "Destriping multisensor imagery with..."

  • ...1(a), there exist eight stripe lines per ten lines, including six successive stripe lines with different variances....

    [...]

References
More filters
Journal ArticleDOI
TL;DR: Methods are presented for obtaining the required information directly from the statistics of the sensor outputs for destriping of LANDSAT Multispectral Scanner images, applying to images obtained with any multisensor line-scan camera.

222 citations


"Destriping multisensor imagery with..." refers background or methods in this paper

  • ...Horn and Woodham (1979) matched the histogram of each subscene to the histogram of the overall image. Brie y, a histogram of the relative frequency of DNs is generated for each sensor and for the overall image. Next, cumulative probability functions (CPFs) are generated from these histograms. Using the overall CPF (i.e. the CPF generated from the overall image) as a reference function, each sensor’s CPF is adjusted. For a given pixel, a cumulative probability value for its original DN is found from its sensor CPF, and that probability value on the reference CPF determines a corrected DN. In practice, the new value is not likely to be an integer and some rule needs to be adopted to select the lower or higher value of the range. This results in a slight (expected value 0.5 DN) change in the average image brightness, well within the noise levels of satellite imagery, but the bias it introduces may have implications in some applications, such as the estimation of atmospheric states using dark targets or in multispectral classiŽ cation (since the bias will by chance be larger in some ranges of some sensor lines). We describe these errors as ‘binning errors’. SpeciŽ cally, the quantization of a continuous variable such as radiance into discrete values is analogous to sorting these values into bins, each bin representing a range of continuous values. KieŒer et al. (1985) showed that about 20% of Landsat-5 TM pixels are mis-binned to an adjacent level by the on-board analogue-to-digita l converter. The histogram equalization technique will introduce binning errors of 1 DN to half the pixels, on average, meaning that after histogram matching we could expect 10% of pixels to diŒer from their ‘correct’ value by 2 DN as a result of binning errors alone. Using the overall image as the reference CPF has the eŒect of increasing the spread of each sensor’s corrected values, since the overall image contains both withinsensor and between-sensor variance. If the striping eŒect is small in comparison to the scene variance, this may not result in visible eŒects, but it may exacerbate binning errors. Increasing the spread of a sensor histogram will result in some empty bins; a result that is unsatisfying when considering a response to a continuous variable like radiance. A particular sensor that diŒers systematically from the others, say in having a much larger oŒset or a larger tendency to fail and record a saturation value, has the potential to distort the overall histogram considerably. Using sensors with similar histograms to construct a reference histogram has been shown to improve destriping (Poros and Peterson 1985). However, matching to a particular ‘typical’ sensor is a simpler solution, and it entirely avoids contributions of sensorto-sensor variance. Algazi and Ford (1981) suggested using a reference sensor to destripe Landsat MSS imagery, but used an arbitrary sensor. To destripe GOES imagery, Weinreb et al. (1989) selected a reference sensor that was stable, low-noise, and did not show clipping at high or low DNs. Wegener (1990) noted histogram matching as proposed by Horn and Woodham (1979) can be sensitive to rather small violations of the similarity assumption. Typically, Landsat MSS band 7 shows a strong bimodality in scenes of mixed land and water. In a sample scene, Wegener (1990) showed that a diŒerence of less than 1% in the number of a single sensor’s pixels sampling water resulted in unsatisfactory destriping....

    [...]

  • ...Horn and Woodham (1979) matched the histogram of each subscene to the histogram of the overall image. Brie y, a histogram of the relative frequency of DNs is generated for each sensor and for the overall image. Next, cumulative probability functions (CPFs) are generated from these histograms. Using the overall CPF (i.e. the CPF generated from the overall image) as a reference function, each sensor’s CPF is adjusted. For a given pixel, a cumulative probability value for its original DN is found from its sensor CPF, and that probability value on the reference CPF determines a corrected DN. In practice, the new value is not likely to be an integer and some rule needs to be adopted to select the lower or higher value of the range. This results in a slight (expected value 0.5 DN) change in the average image brightness, well within the noise levels of satellite imagery, but the bias it introduces may have implications in some applications, such as the estimation of atmospheric states using dark targets or in multispectral classiŽ cation (since the bias will by chance be larger in some ranges of some sensor lines). We describe these errors as ‘binning errors’. SpeciŽ cally, the quantization of a continuous variable such as radiance into discrete values is analogous to sorting these values into bins, each bin representing a range of continuous values. KieŒer et al. (1985) showed that about 20% of Landsat-5 TM pixels are mis-binned to an adjacent level by the on-board analogue-to-digita l converter. The histogram equalization technique will introduce binning errors of 1 DN to half the pixels, on average, meaning that after histogram matching we could expect 10% of pixels to diŒer from their ‘correct’ value by 2 DN as a result of binning errors alone. Using the overall image as the reference CPF has the eŒect of increasing the spread of each sensor’s corrected values, since the overall image contains both withinsensor and between-sensor variance. If the striping eŒect is small in comparison to the scene variance, this may not result in visible eŒects, but it may exacerbate binning errors. Increasing the spread of a sensor histogram will result in some empty bins; a result that is unsatisfying when considering a response to a continuous variable like radiance. A particular sensor that diŒers systematically from the others, say in having a much larger oŒset or a larger tendency to fail and record a saturation value, has the potential to distort the overall histogram considerably. Using sensors with similar histograms to construct a reference histogram has been shown to improve destriping (Poros and Peterson 1985). However, matching to a particular ‘typical’ sensor is a simpler solution, and it entirely avoids contributions of sensorto-sensor variance. Algazi and Ford (1981) suggested using a reference sensor to destripe Landsat MSS imagery, but used an arbitrary sensor. To destripe GOES imagery, Weinreb et al. (1989) selected a reference sensor that was stable, low-noise, and did not show clipping at high or low DNs. Wegener (1990) noted histogram matching as proposed by Horn and Woodham (1979) can be sensitive to rather small violations of the similarity assumption....

    [...]

  • ...Horn and Woodham (1979) matched the histogram of each subscene to the histogram of the overall image. Brie y, a histogram of the relative frequency of DNs is generated for each sensor and for the overall image. Next, cumulative probability functions (CPFs) are generated from these histograms. Using the overall CPF (i.e. the CPF generated from the overall image) as a reference function, each sensor’s CPF is adjusted. For a given pixel, a cumulative probability value for its original DN is found from its sensor CPF, and that probability value on the reference CPF determines a corrected DN. In practice, the new value is not likely to be an integer and some rule needs to be adopted to select the lower or higher value of the range. This results in a slight (expected value 0.5 DN) change in the average image brightness, well within the noise levels of satellite imagery, but the bias it introduces may have implications in some applications, such as the estimation of atmospheric states using dark targets or in multispectral classiŽ cation (since the bias will by chance be larger in some ranges of some sensor lines). We describe these errors as ‘binning errors’. SpeciŽ cally, the quantization of a continuous variable such as radiance into discrete values is analogous to sorting these values into bins, each bin representing a range of continuous values. KieŒer et al. (1985) showed that about 20% of Landsat-5 TM pixels are mis-binned to an adjacent level by the on-board analogue-to-digita l converter....

    [...]

  • ...Horn and Woodham (1979) matched the histogram of each subscene to the histogram of the overall image. Brie y, a histogram of the relative frequency of DNs is generated for each sensor and for the overall image. Next, cumulative probability functions (CPFs) are generated from these histograms. Using the overall CPF (i.e. the CPF generated from the overall image) as a reference function, each sensor’s CPF is adjusted. For a given pixel, a cumulative probability value for its original DN is found from its sensor CPF, and that probability value on the reference CPF determines a corrected DN. In practice, the new value is not likely to be an integer and some rule needs to be adopted to select the lower or higher value of the range. This results in a slight (expected value 0.5 DN) change in the average image brightness, well within the noise levels of satellite imagery, but the bias it introduces may have implications in some applications, such as the estimation of atmospheric states using dark targets or in multispectral classiŽ cation (since the bias will by chance be larger in some ranges of some sensor lines). We describe these errors as ‘binning errors’. SpeciŽ cally, the quantization of a continuous variable such as radiance into discrete values is analogous to sorting these values into bins, each bin representing a range of continuous values. KieŒer et al. (1985) showed that about 20% of Landsat-5 TM pixels are mis-binned to an adjacent level by the on-board analogue-to-digita l converter. The histogram equalization technique will introduce binning errors of 1 DN to half the pixels, on average, meaning that after histogram matching we could expect 10% of pixels to diŒer from their ‘correct’ value by 2 DN as a result of binning errors alone. Using the overall image as the reference CPF has the eŒect of increasing the spread of each sensor’s corrected values, since the overall image contains both withinsensor and between-sensor variance. If the striping eŒect is small in comparison to the scene variance, this may not result in visible eŒects, but it may exacerbate binning errors. Increasing the spread of a sensor histogram will result in some empty bins; a result that is unsatisfying when considering a response to a continuous variable like radiance. A particular sensor that diŒers systematically from the others, say in having a much larger oŒset or a larger tendency to fail and record a saturation value, has the potential to distort the overall histogram considerably. Using sensors with similar histograms to construct a reference histogram has been shown to improve destriping (Poros and Peterson 1985). However, matching to a particular ‘typical’ sensor is a simpler solution, and it entirely avoids contributions of sensorto-sensor variance. Algazi and Ford (1981) suggested using a reference sensor to destripe Landsat MSS imagery, but used an arbitrary sensor....

    [...]

  • ...Horn and Woodham (1979) matched the histogram of each subscene to the histogram of the overall image....

    [...]

Journal ArticleDOI
TL;DR: In this article, the authors investigated the effect of sensor stripes in multisensor imagery and found that they can cause subsequent image classification to fail, if not removed properly, by removing the stripes.
Abstract: Sensor stripes evident in multisensor imagery can cause subsequent image classification to fail, if not removed properly. Of the destriping algorithms investigated, the one published by Horn and Wo...

139 citations


"Destriping multisensor imagery with..." refers background or methods in this paper

  • ...To destripe GOES imagery, Weinreb et al. (1989) selected a reference sensor that was stable, low-noise, and did not show clipping at high or low DNs. Wegener (1990) noted histogram matching as proposed by Horn and Woodham (1979) can be sensitive to rather small violations of the similarity…...

    [...]

  • ...To destripe GOES imagery, Weinreb et al. (1989) selected a reference sensor that was stable, low-noise, and did not show clipping at high or low DNs. Wegener (1990) noted histogram matching as proposed by Horn and Woodham (1979) can be sensitive to rather small violations of the similarity assumption....

    [...]

  • ...Wegener (1990) found moment matching produced a noisy, banded image, but the commercial algorithm he used compared data only within a particular sensor sweep (i.e. within six lines in the MSS imagery used), which dramatically reduced the number of pixels per sensor....

    [...]

  • ...In a sample scene, Wegener (1990) showed that a diŒerence of less than 1% in the number of a single sensor’s pixels sampling water resulted in unsatisfactory destriping....

    [...]

  • ...coastline runs near-parallel to a scan line), the problem can be solved by using a larger image, or alternatively, by segmenting the image, in an extension of Wegener’s (1990) approach....

    [...]

Journal Article
TL;DR: A simple method for the cosmetic removal of scan-line noise from geometrically corrected Landsat Thematic Mapper data is presented and the possible effects upon image signal are discussed.
Abstract: A simple method for the cosmetic removal of scan-line noise from geometrically corrected Landsat Thematic Mapper data is presented. The method uses only standard spatial filters and arithmetic routines that are already present on most image processing systems. Examples are provided, and the possible effects upon image signal are discussed.

134 citations


"Destriping multisensor imagery with..." refers background in this paper

  • ...Since the variation is periodic, lters have been constructed which remove features at a given frequency (Srinivasan et al. 1988, Crippen 1989)....

    [...]

Journal ArticleDOI
TL;DR: A study of EDF matching with data from GOES-7 removed stripes from the image because the technique was used to generate a normalization look-up table from data taken on 18 May 1988, and the table was applied to image data obtained 2 weeks later, on 1 June 1988.

87 citations


"Destriping multisensor imagery with..." refers background in this paper

  • ...To destripe GOES imagery, Weinreb et al. (1989) selected a reference sensor that was stable, low-noise, and did not show clipping at high or low DNs. Wegener (1990) noted histogram matching as proposed by Horn and Woodham (1979) can be sensitive to rather small violations of the similarity…...

    [...]

Journal ArticleDOI
TL;DR: To decrease the dimensionality of the Landsat-4 data, principal component transformation of the data to four significant new bands was performed, and the results compared with latest available land use maps.
Abstract: Techniques have been developed or improved to calibrate, repair, geometrically correct, and extract information from Landsat-4 satellite data. Statistical techniques to correct data radiometry have been evaluated and have minimized striping and banding. It is shown that unless these statistical techniques are used, striping will result even with perfect calibration parameters. Algorithms have been developed to replace data from failed detectors and to reduce coherent noise. The Landsat-4 data have been geometrically corrected to conform to a 1:100 000 map reference to an accuracy of about 41 m. The data were then recorded onto film, and image products produced that can serve as low-cost accurate map products. To decrease the dimensionality of the Landsat-4 data, principal component transformation of the data to four significant new bands was performed, and the results compared with latest available land use maps. The transformation is useful for land use analysis and in delineating vegetation anomalies which appear to reflect areas underlain by altered serpentinite. A range of image processing systems have been used to process the satellite data, including general purpose, special purpose, and personal computers. These systems are described, along with their processing performance. Index Terms-Digital Image Processing, Thematic Mapper, Multispectral Scanner, Calibration, Geometric Correction, Mapping, Digital Terrain, Enhancement, Noise Removal, Personal Computer, Entropy, Principal Components, Banding, Striping, Information Extraction, Geology, Land Use.

77 citations


"Destriping multisensor imagery with..." refers methods in this paper

  • ...Fischel (1984) used moment matching in combination with calibration lamp data (not normally available to end-users) for Landsat TM optical bands and successfully removed all detector-to-detector striping (except binning errors) over homogeneous areas of water. An earlier implementation of moment matching by Rohde et al. (1978) included a logical error which biased the mean corrected value of sensors with high standard deviations upwards and produced unsatisfactory results. Care should be taken to use a fairly large image so that the assumption of statistically similar subscenes is likely to be met. Moment matching has been found to be unsatisfactory when test images are small. For example, Horn and Woodham (1979) used 364 by 430 pixels in a Landsat MSS image (6 sensors) and found moment matching to be unsatisfactory (some 26 000 pixels per sensor). Wegener (1990) found moment matching produced a noisy, banded image, but the commercial algorithm he used compared data only within a particular sensor sweep (i....

    [...]

  • ...Fischel (1984) used moment matching in combination with calibration lamp data (not normally available to end-users) for Landsat TM optical bands and successfully removed all detector-to-detector striping (except binning errors) over homogeneous areas of water. An earlier implementation of moment matching by Rohde et al. (1978) included a logical error which biased the mean corrected value of sensors with high standard deviations upwards and produced unsatisfactory results....

    [...]

  • ...Fischel (1984) used moment matching in combination with calibration lamp data (not normally available to end-users) for Landsat TM optical bands and successfully removed all detector-to-detector striping (except binning errors) over homogeneous areas of water....

    [...]

  • ...Fischel (1984) used moment matching in combination with calibration lamp data (not normally available to end-users) for Landsat TM optical bands and successfully removed all detector-to-detector striping (except binning errors) over homogeneous areas of water. An earlier implementation of moment matching by Rohde et al. (1978) included a logical error which biased the mean corrected value of sensors with high standard deviations upwards and produced unsatisfactory results. Care should be taken to use a fairly large image so that the assumption of statistically similar subscenes is likely to be met. Moment matching has been found to be unsatisfactory when test images are small. For example, Horn and Woodham (1979) used 364 by 430 pixels in a Landsat MSS image (6 sensors) and found moment matching to be unsatisfactory (some 26 000 pixels per sensor)....

    [...]