# Performance metrics for analysing the performance of various filters

We are working with speckle noise reduction in ultrasound images by using various despeckling filtering techniques. We compare the performance of each filtering technique by using Quality metrics such as mean square error,signal to noise ratio,peak signal to noise ratio,Edge preservation index,Speckle index.

We know that mean square error metric indicates how different the images being compared are. Therefore the lower its value is,the closer the estimated image is to the original image. The peak signal to noise ratio must be high which indicates an improvement in speckle reduction.But when we calculate these metrics, mean square error is high and Peak signal to noise ratio is small. any one please say me why this happens so.

The original image that we have used is shown below

Then we corrupted the original image with 3% noise

The formula that we used to calculate these metrics is given below

The result which we obtain while using median filter is given below

• Where did you get your base image? Where did you get your ground truth. In optics we use a USAF target for this kind of calculation. You should consider a control, or else what you are doing is meaningless... – Mikhail Mar 13 '14 at 5:13

They are getting values like that because they may be reading the image as double while calculating the metrics, you can use im2uint8 to convert the double into uint8 and estimate the metrics. you will get the desired output. Though it does not mean that their values are incorrect.