Column

(No.12) Efficient Processing of Perceived Information --- Some Applications ---

Most information perceived by animals is ignored without processing, especially that of images and sound. With this mechanism, animals eliminate processing costs, which enables them to focus only on urgent signals. It is well known that animals focus on moving objects, ignoring still images. Many of these properties are acquired through training and learning.

Humans also use similar mechanisms in the brain. Our visual system responds unconsciously to vertical, horizontal and slanted edges. To activate this mechanism, training is required. Nobel prize laureate, Dr. Konrad Lorenz introduced an experiment in his book in which a newborn cat is put in a room surrounded by walls with only vertical lines; after that, the cat cannot recognize horizontal lines. The cat runs into walls having only horizontal lines. Humans also follow similar learning processes. Our living environment has more horizontal edges than vertical edges. We therefore have much higher resolution for horizontal lines than vertical lines. Accordingly, some new display systems have differing resolutions between the horizontal and vertical directions, which means screens are wider than they are tall, keeping the same number of pixels in total.

At a noisy party, human voices become hard to identify, but words similar to one's own name can be distinguished; this is known as the "cocktail party effect." This effect is explained by a mechanism called object matching. By responding to a word (object) as a unit, without responding to each phoneme, the signal-to-noise ratio jumps. Here is an example:

Consider the procedure to match the two strings "yamamoto san" and "yasumoto san." If we use a sequential matching procedure, the first two characters match correctly, but the 3rd does not. It is here at the 3rd step that the two strings are found to differ (Fig. 1).

Fig. 1 Sequential matching procedure
Fig. 1 Sequential matching procedure

If the comparison is based on a word, the first words "yamamoto" and "yasumoto" have a 7/8 matching ratio, which is 88% correct. The second words "san" and "san" match perfectly (Fig. 2).

Fig.2 Word matching ratio
Fig.2 Word matching ratio

Figure 3 shows examples of signals with various levels of noise. First row A is pure signals. Second row B includes random noise with zero means. Row C indicates one unit-phase shift of row B. Rows C, D, and E have larger phase shifts.

Fig 3 Signal in noise

Fig 3 Signals with unit size are shown in A, and random noise with 0 means it is added; B. C, D, and E include phase shifts of various values.

The figure on the far right shows the inner product of the original signal and current signal. (example: A × B = 7) As shown, the phase shift reduces the inner product dramatically. If we can keep the phase, the inner product does not change much. This nature is effectively utilized for mobile telephone communications and communications from vessels in space.

Fig.4 Example of embedding a sparse image in a dense image
Fig.4 Example of embedding a sparse image in a dense image

Figure 4 shows another example applied to an image. If known signal A is given, we will prepare known picture B. Here, we suppose only internal people know B. After A and B are merged into one image, and image A shares minor information, A will be almost invisible. This method is called "steganography"; it is easily implemented and imbedded under normal conditions. It is also analogously called an electronic envelope.

The above examples show the importance of phase. Can we intentionally control and utilize this phase information?

Figure 5 shows resultant image C after mask B was placed on top of image A with a relative phase difference ofd. This indicates that a small phase controls images as a whole.

Fig. 5(A) A small phase difference delivers a a very different resultant
Fig. 5(B) A small phase difference delivers a a very different resultant

Fig. 5(C) A small phase difference delivers a a very different resultant

If figure B is placed on top of image A, the apparent image changes depending on the relative position (phase) between A and B. Figure C shows two examples that depend on phase δ.

Fig. 5 A small phase difference delivers a very different resultant

Figure 6 is an implementation for a digital thermometer. Here, the phase shift unit of δ = 10μm was used. Materials include plastic and glass with a large expansion coefficient difference. No special energy source is required for this digital display. The thermometer has already been working for ten years. Additional artifices are listed below:

    • Stable expansion coefficient over a long time period.
    • A 2-digit display covering 50 degrees.
    • Overlapped image reduction mechanism of two adjacent figures when the phase difference is less than one unit (δ = 10μm).
    • Adjustable mechanism for maintenance.

Fig. 6 Digital thermometer using phase difference control(conceptual diagram)

Fig. 6 Digital thermometer using phase difference control

Two plates with different thermal expansion rates are place adjacent to one another (See the conceptual diagram on the left). A figure pattern is covered by a mask image. Temperature difference is translated as a phase difference between different materials, which causes a pattern change. The picture on the right is the actual thermometer.

Fig. 6 Digital thermometer using phase difference control

Abundant noise exists in our natural world and signals are always in the noise. We can even say that the main subject of statistics is how to handle noise. On the other hand, however, living creatures can deal with noise unconsciously. These phenomena give us significant hints to open up new and innovative applications.
(Ej,2004.06)



Back to index

Page Top