III. EXPRIMENTAL RESULTS AND DISCUSSION
In this section we performed several experiments to test the
performance of the proposed method using various test images
and noisy medical ultrasound images.
To quantify the performance of a speckle noise reduction
filtering algorithms in terms of efficiency and enhancing the
significant image information, we used several quality evaluation
metrics such as average difference(AD), mean square
error(MSE), peak signal to noise ratio(PSNR), maximum
difference(MD), normalized cross-correlation (NK), and structural
content(SC) [24]. All the metrics are self and separate
explanation for each and every metrics is not included in this
discussion due to page limitation.
The proposed method is tested on the Lena and geometric
images corrupted by a speckle noise, with a variance equal
to 0.01 (Figure III.1b and III.2b). In order to demonstrate
its effectiveness, we compare our algorithm to the anisotropic
diffusion algorithm. The evaluation measurements were done
for each despeckled image. Figure III.1 and III.2 shows representative
filtered results of synthetic images. The proposed
algorithm is further tested on real clinical ultrasound images
as shown in figure III.3 and III.4. Furthermore, the correspondence
to various quality evaluation metrics are calculated
for the filtered images and shown in data Table I and II.
From Figure III.1 and III.2, we can observe that the speckle
noise has been greatly alleviated with the proposed method,
which produces an obvious combination of speckle reduction,
smoothing, and features enhancing. Therefore, comparative
results are shown in figure III.3 and III.4 clearly indicate that
our algorithm presents a superior edge preserving behavior
with an improvement of contrast enhancement, and its filtering
results are the best visual appearance as compared to the
anisotropic diffusion. The results of this study showed that
the anisotropic diffusion filter was not able to enhance image
quality and produced blurred edges in the filtered image (see
figure III.3 and III.4 ). Referring to Table I and II, the PSNR
of the proposed filter are larger than the anisotropic filter in all
experiments, and their metrics of MSE, NK, AD, SC and MD
are smaller. Although the interval scale between all metrics
of the filters is not large, it is worthwhile to note that the
mathematical expressions of MSE, PSNR, AD, NK, SC and
MD are rather significant.
The obtained results confirm that the proposed method gives
a very good performance : (i) in terms of smoothing and at
the same time edge preservation without any blurring or distortions
near the edges(ii) in terms of contrast enhancement (iii)
Quantitatively it gives best image quality evaluation results
compared to anisotropic diffusion filter, which confirms that
our method seems to be more suitable for speckle reduction
in ultrasound images and can achieve better performance than
anisotropic diffusion algorithm.
(a) Original image (b) Noised image
(c) Anisotropic diffusion (d) Proposed method
Figure III.1: Experimental results on Lena image
Metrics MSE PSNR NK AD SC MD
Proposed filter 0.0025 74.1964 0.9870 -2.7807e-016 1.0173 0.3980
Anisotrope filter 0.0032 73.0811 0.9632 0.0064 1.0580 0.4762
Table I: Image quality evaluation metrics of Lena image
سوم تجربي نتایج و بحثدر این بخش ما در انجام آزمایش های مختلف برای تستعملکرد روش پیشنهادی با استفاده از تصاویر تست مختلفو تصاویر سونوگرافی پزشکی پر سر و صدا.به راندمان لکه کاهش سر و صدافیلتر کردن الگوریتم از نظر بهره وری و بالا بردناطلاعات تصویر قابل توجهی به ما استفاده می شود چندین کیفیت ارزیابیمعیارهای مانند difference(AD) متوسط ميانگين مربعاتerror(MSE)، پهناي قله ratio(PSNR)، سر و صدا حداکثرdifference(MD) همبستگی نرمال (NK) و سازهcontent(SC) [24]. تمام معیارهای خود و جداتوضیح برای هر متریک در این وجود داردبحث با توجه به محدودیت صفحه است.روش تست شده است در لنا و هندسیتصاویر خراب شده با نویز لکه با برابر واريانسبه 0.01 (شکل III.1b و III.2b). به منظور نشان دادناثر آن به ما مقایسه الگوریتم ما به ناهمسانگردالگوریتم انتشار. اندازه ارزیابیبرای هر تصویر despeckled. شکل III.1 و III.2 نماینده دهدفیلتر نتایج تصاویر مصنوعی. پیشنهادیالگوریتم است که بیشتر در عکس سونوگرافی واقعی بالینی آزمایشهمانطور که در شکل نشان داده شده III.3 و III.4. علاوه بر این، مکاتباتمتریک برای ارزیابی کیفیت مختلف محاسبه می شودبرای فیلتر تصاویر و داده های نشان داده شده در جدول اول و دوم.از شکل III.1 و III.2 که مشاهده می کنیم می توانید لکهnoise has been greatly alleviated with the proposed method,which produces an obvious combination of speckle reduction,smoothing, and features enhancing. Therefore, comparativeresults are shown in figure III.3 and III.4 clearly indicate thatour algorithm presents a superior edge preserving behaviorwith an improvement of contrast enhancement, and its filteringresults are the best visual appearance as compared to theanisotropic diffusion. The results of this study showed thatthe anisotropic diffusion filter was not able to enhance imagequality and produced blurred edges in the filtered image (seefigure III.3 and III.4 ). Referring to Table I and II, the PSNRof the proposed filter are larger than the anisotropic filter in allexperiments, and their metrics of MSE, NK, AD, SC and MDare smaller. Although the interval scale between all metricsof the filters is not large, it is worthwhile to note that themathematical expressions of MSE, PSNR, AD, NK, SC andMD are rather significant.The obtained results confirm that the proposed method givesa very good performance : (i) in terms of smoothing and atthe same time edge preservation without any blurring or distortionsnear the edges(ii) in terms of contrast enhancement (iii)Quantitatively it gives best image quality evaluation resultscompared to anisotropic diffusion filter, which confirms thatour method seems to be more suitable for speckle reductionin ultrasound images and can achieve better performance thananisotropic diffusion algorithm.(a) Original image (b) Noised image(c) Anisotropic diffusion (d) Proposed methodFigure III.1: Experimental results on Lena imageMetrics MSE PSNR NK AD SC MDProposed filter 0.0025 74.1964 0.9870 -2.7807e-016 1.0173 0.3980Anisotrope filter 0.0032 73.0811 0.9632 0.0064 1.0580 0.4762Table I: Image quality evaluation metrics of Lena image
ترجمه، لطفا صبر کنید ..
