Vol.2 No.4 2010

10/68

Research paper : Development of a real-time all-in-focus microscope (K. Ohba)−240−Synthesiology - English edition Vol.2 No.4 (2010) High processing capacity was required for the Laplacian filter and smoothing filters in the preliminary processing of the image in this method. Figures 10 and 11 show the hardware version of this process, which is an example of the vision chip implemented in Prototype 1 that will be explained later in this chapter. For the hardware used, the chip allows easy SIMD processing, and therefore, to calculate the above IQM value, as a process using the pixel values around each pixel points, parallel calculation of the Laplacian and smoothing filters are necessary. The examples of SIMD processing for each operation are shown. Looking at each calculation result, Laplacian filter (R12 of Fig. 10) and smoothing (R5 of Fig. 11) were ultimately obtained for each pixel point.For the final product realization, while using this processing method as a reference, we configured a system that calculated the IQM value using the FPGA. The external appearance of Prototype 1 system created in 1999 is shown in Fig. 12.As an assessment test, we conducted processing by placing an object with a depth of 35 mm at a position of 160 mm from the varifocal lens. To move the focal distance to cover an object with a height of 35 mm, 21 images were captured, and the resolution of depth was 1.67 mm. While the spatial resolution of the system is dependent on the setting of the optical device, in this case, since 16 mm × 16 mm was processed at an image resolution of 256 × 256, the resolution was 62.5 m/pixel.The object used here was an artificial four-step pyramid of a height of 35 mm. The first step was 10 mm and a height of 10mm, the second step was 7 mm and a height of 10 mm, the third step was 4 mm and a height of 10 mm, and the fourth step was 3 mm and a height of 5 mm.Part of the 21 images shot while moving the focal distance is shown in Fig. 13 on the left. As mentioned earlier, by conducting processing, the all-in-focus image as in Fig. 13 on the right and the VR display as in Fig. 14 were obtained. While the all-in-focus image itself was adequate, we obtained a largely dispersed result since the depth image had a small region that required smoothing, and the resolution in the depth direction was small.While the performance of Prototype 1 was dependent on the Fig. 10 Laplacian filter processing operation by SIMD processing.Fig. 12 Prototype 1.Fig. 11 Summation processing operation by SIMD processing.R1R2R2R30255R4=R2-R1R6=R5-R4R9=R7-R2R8=LOR(R2)R7=ROR(R2)R10=R8-R2R11=R9+R10R12=R6+R11R5=R3-R2I(i, j-1)-I(i-1, j-1)I(i, j-2)-I(i, j-1)I(i, j-1)-I(i, j)I(i, j+1)-I(i, j)I(i, j)-I(i, j+1)I(i, j+2)-I(i, j+1)I(i, j)-I(i, j-1)I(i, j+1)-I(i-1, j+1)I(i, j)-I(i-1, j)I(i+1, j-1)-I(i, j-1)I(i+1, j+1)+I(i-1, j+1)-2I(i, j+1)I(i+1, j-1)+I(i-1, j-1)-2I(i, j-1)I(i+1, j)+I(i-1, j)-2I(i, j)I(i+1, j+1)+I(i-1, j+1)+I(i, j+2)+I(i, j)-4I(i, j+1)I(i+1, j-1)+I(i-1, j-1)+(i, j)+I(i, j-2)-4I(i, j-1)I(i-1, j)+I(i+1, j)+I(i, j+1)+I(i, j-1)-4I(i, j)I(i, j+2)+I(i, j)-2I(i, j+1)I(i, j)+I(i, j-2)-2I(i, j-1)I(i, j+1)+I(i, j-1)-2I(i, j)I(i+1, j+1)-I(i, j+1)I(i+1, j)-I(i, j)I(i-1, j-1)I(i+1, j-1)I(i-1, j)I(i-1, j+1)I(i, j+1)I(i+1, j+1)I(i+1, j)I(i, j-1)I(i, j-1)I(i, j-2)I(i, j-1)I(i, j+1)I(i, j+1)I(i, j+2)I(i, j)I(i, j)I(i, j)I(i, j)alkjhgefidcbalkjhgefidcbalkjhgefidcbalkjhgefidcb010101010101000000akgeicalkjhgefidcb010101010101ljhfdb000000alkjhgefidcb010101010101010101010101010100110000a+ba+b+c+da+ba+ba+bc+dc+dc+dc+de+fe+fe+fe+f+g+he+fg+hg+hg+hg+hi+ji+ji+ji+jk+lk+lk+l+i+jk+lk+lMASK(1)MASK(1)MASK(1)MASK(2)MASK(3)R1R4=cros(R3, 2)R2=cros(R1, 1)R1R3=R1+R2R5=R3+R4R2=ROR(R1)R4=ROL(R1)R5=R4&MASK(1)R3=R2&MASE(1,R)R6=R3 or R5cros(R1, 1)

※このページを正しく表示するにはFlashPlayer9以上が必要です