Experimental Results - kovimesterr/SSIP2013 GitHub Wiki

To evaluate the performance of a technique, we may ask an expert or use some standard criteria. For our project, we had no access to an expert so we searched for some suitable measurement. These values also allow us to compare different methods performance. In this section, we start by explaining three well known criteria, available in the literature and then we present the result of the all techniques on 40 images from the Mammographic Image Analysis Society (MIAS) database were selected [6].

where Imax and Imin are the the maximal and minimal intensities of the image, and If, Ib are the minimal and maximal intensities of the foreground and the background image. The Michelson contrast measure is used to measure the contrast of a periodic pattern such as a sinusoidal grating, while the Weber contrast measure assumes a large uniform luminance background with a small test target. Both measures are therefore unsuitable for measuring the contrast in complex images.

Additionally, objective image quality measure play important roles in various image processing applications. There are basically two classes of objective quality or distortion assessment approaches. The first are mathematically defined measures such as the widely used mean squared error (MSE), peak signal to noise ration (PSNR), root mean squared error (RMSE), mean absolute error (MAE), and signal to noise ration (SNR). The second class of measurement methods consider human visual system characteristics in an attempt to incorporate perceptual quality measures. Unfortunately, none of these complicated objective metrics in the literature has shown any clear advantage over simple mathematical measures such as RMSE and PSNR under strict testing conditions and different image distortion environments.

Mathematically defined measures are still attractive because of two reasons. First, they are easy to calculate and usually have low computational complexity. Second, they are independent of viewing conditions and individual observers. Although, it is believed that the viewing conditions play important roles in human perception of image quality, they are, in most cases, not fixed and specific data is generally unavailable to the image analysis system. If there are N different viewing conditions, a viewing-dependent method will generate N different measurement results that are inconvenient to use. In addition, it becomes the user’s responsibilities to measure the viewing conditions and to calculate and input the condition parameters to the measurement systems. By contrast, a viewing condition independent measure delivers a single quality value that gives a general idea of how good the image.

We reviewed the measurement techniques in the literature [1][2] and we found four types of frequently used quality measurements: Absolute Mead Brightness Error (AMBE), Measure of Enhancement by Entropy (EME), Discrete Entropy (DE), Universal Image Quality Index (UIQ).

AMBE

One way of measuring the enhancement level is the AMBE technique [2]. AMBE is defined as the absolute difference between the input and output mean. The expression for AMBE may be given as

Where E (X) is the mean of the input image, E (Y) is the mean of the output image. A median value of AMBE implies better brightness preservation. Either a very low value or the highest value of AMBE also indicates poor performance in case of contrast enhancement.

EME

The concept of local intensity measurement is based on Weber law, which argued that the human visual interpretation depends on the ratio of light intensity values. This quantitative measure is also useful in choosing enhancement methods, parameters in parametric enhancement methods. Let the image be split into k1k2 blocks wk,l(i, j). Then the enhancement performance measure (EME) [4] [5] is calculated as follows:

where Iwmax and Iwmin are respectively respectively maximum and minimum of the image block wk,l(i, j), after processing the block by an enhancement algorithm.

When the value of EME is too high, it indicates over enhancement in the output image and it shows a loss of local information due to washed-out output image or leads to insufficient medical details during diagnosis and sometimes it situation introduces artifacts in the resultant image. On the other hand, a very low value of EME indicates hidden information is not significantly enhanced. Then it is necessary to have an optimum value of EME in order to have both contrast enhancement and preserving more local details of the mammogram images.

Entropy

Entropy is a well-known statistical measure of randomness that can be used to characterize the texture of the input image. Entropy is defined as

where p contains the histogram of the image.

UIQ

This quality index models any distortion as a combination of three different factors: loss of correlation, luminance distortion, and contrast distortion. Let x=(xi) i=1,2,...,N and y=(yi) i=1,2,...,N be the original and the enhanced image signals, then UIQ [3] can be defined as follows

where sigmax, sigmay denote the standard deviation of the original image x,y and

The first component in UIQ is the correlation coefficient, the second measures how close the mean luminance is between x and y. The third one measures how similar the contrasts of the images are. The range of values of UIQ is [0,1], where the best value 1 is achieved if and only if sigmax=sigmay.

Results

In our tests we used the mini-MIAS database of mammograms [7]. This is a very large database which means 322 recordings with 1024x1024 resolution. Every image is annotated, and the center point of abnormalities have been stored. Furthermore, we chose 39 images related to cancer and normal cases. We considered different types of breast with dense and fatty tissue as well.

Besides above techniques, we also tried to evaluate the performance of following contrast enhancement methods in MATLAB image processing toolbox in order to enrich the comparison results and eliminate any chance of overlooking some currently available efficient techniques:

Decorrstretch(): Decorrelation stretching is a way to enhance the color differences in an image and therefor it can be used to improve the contrast of a mammogram.

Imadjust(): similar to Decorrelation stretching, this function also modifies the values in grayscale image in order to increases the contrast of the output image such that 1% of data is saturated at low and intensities of input image.

Imsharpen(): returns an enhanced version of the input image, where the image features, such as edges, have been sharpened using the unsharp masking method. The algorithm, therefore, can reveal the hidden structure of the breast tissue.

Following graphs represent the final results with respect to the presented quality measures.

Conclusions

1. The results are straightforward in the case of UIQ because it is an absolute measure. Here higher values represents better score. CLAHE, BDEC, h-dome methods have very low performance comparing, but the others are more or less the same.
2. The entropy of these methods are very similar to the original image, except the previously mentioned three methods. In the case of CLAHE, h-dome algorithms the entropy is much more less than the original which means we lose a lot of information. BDEC has higher entropy, but it has very bad UIQ. By this reason we conclude that new, non-existing information has been generated by this method.
3. The AMBEs measurements should have optimum values to preserve the naturalness of the original mammogram image. The imsharpen and the imadjust, h-dome, CLAHE procedures have very low and very high AMBE which means poor performance in this sense.
4. For EME h-dome and CLAHE methods have bad results again which means they are over enhance the images. On the other hand in case of contrast enhancement imsharpen and the imadjust have better performance. For the rest of these algorithm the rate of quality improvement is insignificant.

Summarizing the results, CLAHE, BDEC and h-dome methods are the worst algorithms for contrast enhancement for these type of imges. The remaining procedures are quiet good in the sense of enhancement, and they are capable to preserve the relevant information of the mammograms.

References

• [1] Gy. Kovács, A. Fazekas, Automatikus gamma korrekció, Proceedings of 8th Conference of the Hungarian Association for Image Processing and Pattern Recognition, Szeged, Hungary.
• [2] M. Sundaram, K. Ramar, N. Arumugam, G. Prabin, Histogram Modified Local Contrast Enhancement for Mammogram Images, Applied Soft Computing, vol. 11, 2011, pp. 5809-5816.
• [3] Z. Wang, A. C. Bovik, An Universal Image Quality Index, IEEE Signal Processing Letters, vol. 20, 2002, pp. 1-4.
• [4] A. Kanojia, S. A. Sos, P. Karen, New Contrast Measure for Transform Based Image Enhancement.
• [5] B. Silver, S. A. Sos, P. Karen, Contrast Entropy Based Image Enhancement and Logarithmic Transformation Coefficient Histogram Shifting.
• [6] http://www.mammoimage.org/databases/
• [7] J. Suckling, The Mammographic Image Analysis Society Digital Mammogram Database, Exerpta Medica. Internatial Congress Series, 1994, pp. 375-378.