Table 3 Comparison between conventional preprocessing methods.
From: Automatic enhancement preprocessing for segmentation of low quality cell images
Methods | Mouse liver cell image dataset | iRPE cell image dataset | |||||
|---|---|---|---|---|---|---|---|
Average | Cytoplasm | Membrane | Nucleus | Average | Background | Membrane | |
w/o preprocessing | 57.90±1.33 | 71.38±4.80 | 40.32±3.84 | 62.01±2.65 | 62.86±0.37 | 76.09±0.09 | 49.64±0.69 |
Median filter (kernel size=3) | 57.94±1.72 | 72.55±3.36 | 39.53±2.81 | 61.74±3.76 | 58.10±0.46 | 73.76±0.21 | 42.44±0.93 |
Median filter (kernel size=9) | 50.01±2.29 | 67.59±5.25 | 32.15±4.22 | 50.29±4.56 | 39.38±0.16 | 62.43±1.18 | 16.33±1.30 |
Gaussian filter (kernel size=3) | 57.62±1.17 | 71.79±4.07 | 40.18±3.95 | 60.90±1.72 | 59.50±0.23 | 74.67±0.18 | 44.33±0.32 |
Gaussian filter (kernel size=9) | 56.32±1.80 | 71.12±3.96 | 38.81±2.62 | 59.04±3.91 | 49.80±1.26 | 69.10±0.74 | 30.50±3.18 |
Bilateral filter (kernel size=3) | 58.38±1.40 | 72.60±3.90 | 40.66±3.47 | 61.86±2.79 | 62.88±0.39 | 75.81±0.31 | 49.95±0.49 |
Bilateral filter (kernel size=9) | 58.41±1.29 | 72.22±4.34 | 40.75±4.10 | 62.90±2.49 | 62.90±0.39 | 76.05±0.06 | 49.74±0.76 |
Sobel filter (kernel size=3) | 55.13±1.57 | 70.18±4.42 | 37.62±2.67 | 57.59±2.65 | 50.22±0.64 | 72.42±0.12 | 28.02±1.39 |
Sobel filter (kernel size=9) | 55.84±1.95 | 70.04±3.52 | 38.62±2.77 | 58.87±3.05 | 34.93±0.12 | 69.86±0.25 | 0.00±0.00 |
Frangi/vesselness filter | 53.09±0.18 | 69.43±1.38 | 32.59±2.41 | 57.24±3.79 | 55.27±2.29 | 72.83±0.93 | 37.71±2.76 |
AEP | 59.27±1.81 | 72.40±3.92 | 42.44±2.27 | 62.98±3.57 | 63.81±0.39 | 75.98±0.32 | 51.64±0.71 |
