Anna Linderhed

2D empirical mode decompositions in the spirit of image compression

The Empirical mode decomposition (EMD) is an adaptive decomposition of the data, as is the Wavelet packet best basis decomposition. This work present the first attempt to examining the use of EMD for image compression purposes. The Intrinsic Mode Function (IMF) and their Hilbert spectra are compared to the wavelet basis and the wavelet packet decompositions expanded in each of its best bases on the same data. By decomposing the signal into basis functions, the waveforms in the signal is represented by the basis and a set of decorrelated discrete values in a vector. A coding scheme is presented where the idea is to decompose the signal into its IMF:s where only the max and min values for each IMF is transmitted. The reconstruction of the IMF in the decoder is done with spline interpolation. We have in the two-dimensional EMD an adaptive image decomposition without the limitations from filter kernels or cost functions. The IMF:s are, in the two-dimensional case, to be seen as spatial frequency subbands, with various center frequency and bandwidth along the image.

VARIABLE SAMPLING OF THE EMPIRICAL MODE DECOMPOSITION OF TWO-DIMENSIONAL SIGNALS

Previous work on empirical mode decomposition in two dimensions typically generates a residue with many extrema points. In this paper we propose an improved method to decompose an image into a number of intrinsic mode functions and a residue image with a minimum number of extrema points. We further propose a method for the variable sampling of the two-dimensional empirical mode decomposition. Since traditional frequency concept is not applicable in this work, we introduce the concept of empiquency, shortform for empirical mode frequency, to describe the signal oscillations. The very special properties of the intrinsic mode functions are used for variable sampling in order to reduce the number of parameters to represent the image. This is done blockwise using the occurrence of extrema points of the intrinsic mode function to steer the sampling rate of the block. A method of using overlapping 7 × 7 blocks is introduced to overcome blocking artifacts and to further reduce the number of parameters required to represent the image. The results presented here shows that an image can be successfully decomposed into a number of intrinsic mode functions and a residue image with a minimum number of extrema points. The results also show that subsampling offers a way to keep the total number of samples generated by empirical mode decomposition approximately equal to the number of pixels of the original image.

Land mine detection by IR temporal analysis: physical numerical modeling

The overall objective of this paper is to improve the understanding of thermodynamic mechanisms around buried objects. The purpose is to utilize most favourable conditions for detection and also to enhance and evaluate other detection methods shown in a companion paper. This paper focuses on physical based models and simulations with measured data as boundaries for different situations of buried objects. For numerical models some assumptions of the real environment and boundaries have to be made, this paper shows the effects of different approaches of these assumptions. The investigations are carried out using a FEM approach with measured weather data as well as different sub models for the boundaries. All modelling works are carried out very in close connections with experiments with the purpose to achieve high accordance between measured and simulated values. This paper shows experimental and simulated results and discusses also the temporal analysis of thermal IR data.

Optical detection of land mines at FOI

This paper presents activities concerning optical detection of landmines at FOI, former FOA. The work is focused on the understanding of the origin of detectable optical signatures for choosing the most favorable conditions for detection. Measurements in test beds and calculations using a thermodynamic FEM model with conditions similar to those of the measurements are compared and interpreted in order to explain the behavior of the contrast. Examples will be given on modeling of buried landmines in soil. The heat flow as well as moisture flow has been taken into consideration. The diurnal heat exchange between the soil surface and the atmosphere generates the contrasts in the infrared images. Calculated temperature differences between the background and the surface above the buried object are compared to measured data from experiments. Results are presented and show how the temperature differences can vary over a 24-hour period. The variation depends on the weather at the time as well as the weather before the measurements started. Results from processing and analysis of temporal variations of optical signals from buried landmines and backgrounds are presented as well as their relation to weather parameters. A detection approach including the Likelihood Ratio Test (LRT) is presented. Some of the work has been carried out in an international cooperation project, Airborne Minefield Area Reduction (ARC). The objective is to develop, demonstrate and promote a new system for performing the UN Level 2 surveys allowing a quick reduction of suspected mine polluted areas and post cleaning quality control.

Learning boundaries on military operational plans from simulation data

In this paper we learn indicators from simulated data that serve as boundaries on military operational plans of an expeditionary operation. These are boundaries that an operation must not move beyond without risk of drastic failure. We receive simulated and evaluated partial patterns of plan instances from a simulation-based decision support system that are patterns of integer strings. These partial patterns are clustered by an unsupervised neural Potts spin clustering method into clusters where the instances in each cluster have similar characteristics and outcomes. This gives all partial patterns a classification. We use a Dempster-Shafer theory based factor screening method on each pair of clusters, where all activities of the plan are evaluated as to their differentiating capacity between the two sets of partial plan instances. All plan instances are projected from their full integer string representation to a subset of factors with high differentiating capacity. We apply supervised learning by Support Vector Machine using the previous classification to learn support vectors for each pair of clusters given the projected plan instances of these clusters. From these support vectors we derive a lower dimension hyper plane that will serve as one of the indicators. One indicator from each pair of clusters will make up a full set of indicators for this operational plan. This set of indicators can be provided to the intelligence service and used during execution of the plan for assessment of its progress, and serve as a warning bell if the plan approaches an indicator which it should not proceed beyond.

Land mine detection by IR temporal analysis: detection method

The Swedish Defence Research Agency (FOI) has presented several approaches to temporal analysis of thermal IR data in the application of mine detection during the years. Detection by classification is performed using a number of detection algorithms with varying, in general good, results. The FOI temporal analysis method is tested on images randomly chosen from a diurnal sequence. The test sequence show very little contrast. The reference features are taken from a known object in the scene or from a numerical model of the object of interest. In this paper variations of the method are evaluated on the same test data. Focus is on the question if increased number of data collection times affects the detection rate and false alarm rate. The ROC curves show performance better than random for all of the tested cases, and excellent for some. Detection rate increases and false alarm rate decreases with increased number of images used for some of the tested cases.

Temporal method for IR minefield feature detection

The overall objective of this work is to investigate the possibilities of using airborne IR sensors for the purpose of detecting minefield features, such as land mines. A method is proposed for temporal analysis by extracting relevant information from diurnal IR images utilizing a combination of thermodynamic modelling, signal and image processing. This paper presents results from a field test of level 2 survey in May 2003 of suspected mine-polluted areas in Croatia. Airborne data was acquired using an IR sensor mounted on a rotary wing UAV. A weather station was used to collect weather data, and pt-100 temperature sensors recorded the temperature gradient in the soil and in reference markers that were used for calibrating the IR camera. The proposed method compares simulated temporal temperature with image data collected at several times during a diurnal cycle from the same area, pixel by pixel. The images are co-registered and calibrated with respect to reference values. The numerical model is based on physical laws and is set with relevant properties, geometries, materials, surface coefficients and the influence of the actual weather sets the boundary conditions. This paper shows some results from using temporal features for detection of different relevant objects in a real minefield.

Optical measurements of real minefields

This paper presents preliminary analysis of the data from measurements on a minefield in Croatia done in the international cooperation project Airborne Minefield Area Reduction (ARC). Temperature differences above and around suspected mines and minefield indicators, were recorded with a long wave IR camera in 8-9 micrometers , over a time of several days, capturing data under different weather conditions. The data are compared to simulations of land mines, minefield indicators and other objects using a themodynamic FEM model, developed at FOI. Different detection methods are presented and applied to the data.

Infrared animal modeling for training ATR algorithms

There are very good automatic detection algorithms available to be used in an Automatic Target Recognition applications. However they need lots of data for training the detector for the specific use, e.g., performing an inventory of wild animals. Ongoing work use thermally correct infrared models of animals for training the detector because collecting real images from these wild animals is too expensive if even possible. This paper describes the process of designing a good IR model of the animals, and the validation process of the thermal model. Several animals are modeled using RadThermIR to be used for training detection algorithms. Animal models are based on commercially available CAD models and are initiated by temperature values from real IR measurements in several different weather conditions. The modeling extends the available set of training images by introducing different weather conditions and different poses of the animal. Fat and fur thickness of the animal is modeled with respect to climate and weather.

Multi-optical mine detection: results from a field trial

As a part of the Swedish mine detection project MOMS, an initial field trial was conducted at the Swedish EOD and Demining Centre (SWEDEC). The purpose was to collect data on surface-laid mines, UXO, submunitions, IEDs, and background with a variety of optical sensors, for further use in the project. Three terrain types were covered: forest, gravel road, and an area which had recovered after total removal of all vegetation some years before. The sensors used in the field trial included UV, VIS, and NIR sensors as well as thermal, multi-spectral, and hyper-spectral sensors, 3-D laser radar and polarization sensors. Some of the sensors were mounted on an aerial work platform, while others were placed on tripods on the ground. This paper describes the field trial and the presents some initial results obtained from the subsequent analysis.