资源描述
Unit 17 Images and Televisions,Passage A Fundamental Concepts Passage B Compression/Decompression Techniques Passage C Television,Passage A Digital Image Fundamentals 1. Digital Image Resolution A digital image is made up of many rows and columns of pixels. For gray scale images, each pixel is assigned a number that represents the gray shade assigned to that pixel. The larger the number of pixels in an image, and the larger the number of available gray scale levels, the better the resolution of the image. Figure 17.1 is an 17-bit images, with 217256 possible gray scale levels. The number of row and column are 303228.,Figure 17.1 An 30322817 Digital Image,2. Histograms The gray scales present in a digital image can be summarized by its histogram (see Figure 17.2)The histogram reports the number of pixels for each grayscale level present in the image, as a bar graph. When an image uses only a small portion of the available gray scale levels, histogram equalization can be used to spread out the usage of gray scale levels over the entire available range1This procedure re-assigns gray scale levels so that image contrast is improved.,Figure 17.2 A Histogram,3. Addition and Subtraction of Images Digital images can be added and subtracted pixel-by-pixel. Adding two images can combine two sets of objects into a single image. Moreover, several noisy images of the same scene can be averaged together to reduce the effect of noise. Image subtraction, on the other hand, can be used to remove an unwanted background from an image. Subtraction of two time-lapsed photographs will show where motion has occurred between the two.,When two images are added or subtracted, the resultant matrix will frequently contain illegal gray scale values. For example, when a pixel in one 17-bit image has the gray scale level 129 and the corresponding pixel in a second 17-bit image has the gray scale level 201, the sum pixel is 129+201330. This is outside the legal range for an 17-bit image, which may only contain gray scale levels between 0 and 255. When the same two images are subtracted, the difference pixel is 129-201= -72, again a value outside the legal range. For these reasons, scaling follows most image arithmetic. Scaling to the range 0, GSLmax may be accomplished as follows: ,4. Warping and Morphing Warping and morphing are digital image techniques that are finding application not only in entertainment but also in medical imaging. Warping stretches or re-shapes an object in an image, while morphing transforms one image into another. These transformations may be accomplished by marking control points, control lines, or triangles in a source image and choosing their new positions in a destination image. The transition between source and destination images is then accomplished by smoothly transforming not only the control element locations, but also their colors. The locations and colors of pixels not explicitly marked as control elements are determined by the locations and colors of the control elements that are nearest.,5. Image Filtering Digital images can be filtered using two-dimensional convolution with a convolution kernel. When an NN image is filtered by an MM convolution kernel, (M-1/2) rows and columns on each side of the image are lost to boundary effects. Low pass filters blur images, high pass filters emphasize sharp changes in gray scale level, and edge filters locate edges in an image.2,6. Dilation and Erosion Dilation adds a layer of pixels to all objects in an image. Erosion removes one layer of pixels from all objects. When dilation is followed by erosion, gaps in broken boundaries identified through edge detection can be filled in. Conversely, when erosion is followed by dilation, spots of noise in an image are removed. Successfully detecting the edges in an image is the first step towards confident identification of object boundaries and then objects recognition. From boundary information, shape characteristics like perimeter and area can be calculated, which can be used to classify an object.,7. Image Spectra Two-dimensional FFTs are used to analyze the spectra of digital images. Just as in the one-dimensional case, a two-dimensional spectrum comprises a magnitude spectrum and a phase spectrum. The phase spectrum carries the best information about the locations of the objects in the image.3When all magnitudes are set to one, the phases alone still show a facsimile of the original image. When all phases are set to zero, the magnitudes alone show no trace of it. Image spectra form the basis for both CT(computed tomography)and MRI (magnetic resonance imaging)scan displays. CT scans are X-rays taken in many directions in a single plane of an object.4 MRI scans depend instead on the magnetic properties of an object placed in a varying magnetic field. Both types of scans permit non-invasive investigations of three-dimensional objects.,8. Image compression In part due to the Internet, digital images are transmitted from place to place more often than ever. To save time and bandwidth(space), both images and other files are often compressed before being transmitted. Lossless compression means that a file is compacted without losing any information, so that the reconstructed file is identical to the original.5 Lossy compression means that some information from the original file is irretrievably lost, but generally the reconstructed file is extremely close to the original. The compression ratio is the ratio of the original file size to the compressed file size.,One simple compression scheme is run-length encoding, which codes more than three repetitions of a number as three copies of the number followed by a count of the additional copies needed. Another compression scheme is Huffman encoding, which uses shorter codes to represent the most common signal elements, and longer codes to represent the least common signal elements. JPEG, an extremely common image compression scheme, uses the discrete cosine transform (DCT) to concentrate most of the information about an 1717 sub-block of an image into a few significant coefficients.6 It then uses both run-length encoding and Huffman encoding to provide further compression.,Notes 1 When an image uses only a small portion of the available gray scale levels, histogram equalization can be used to spread out the usage of gray scale levels over the entire available range. 当一幅图像只使用了可用灰度级的一小部分时, 可以使用直方图均衡的方法将灰度级的使用扩展到整个可用的范围。 2 Low pass filters blur images, high pass filters emphasize sharp changes in gray scale level, and edge filters locate edges in an image 低通滤波器使图像变得模糊, 而高通滤波器突出了图像的灰度锐变, 边缘滤波器则对图像边缘进行定位。 ,3 The phase spectrum carries the best information about the locations of the objects in the image 相位谱携带着图像中目标位置的信息。 4 CT scans are X-rays taken in many directions in a single plane of an object. CT技术是使用X射线从不同方向对目标的某个平面进行扫描。 5 Lossless compression means that a file is compacted without losing any information, so that the reconstructed file is identical to the original 无损压缩是不损失任何信息地将文件进行压缩, 重建得到的文件和原文件是完全一样的。,6 JPEG, an extremely common image compression scheme, uses the discrete cosine transform (DCT) to concentrate most of the information about an 1717 sub-block of an image into a few significant coefficients. JPEG是一种极其常用的图像压缩方法, 该方法使用离散余弦变换将图像中1717小块的大部分信息集中到少数几个重要系数上。 ,Exercises 1. Please translate the following phrases into English. (1) 灰度图像 (2) 直方图均衡 (3) 图像对比度(4) 结果矩阵 (5) 边缘检测(6) 行程编码 (7) 无损压缩(8) 低通滤波器 (9) 哈夫曼编码(10) 数字图像压缩,2. Please translate the following phrases into Chinese. (1) edge filter (2) magnitude spectrum (3) object recognition (4) phase spectrum (5) CT(Computed Tomography) (6) MRI(Magnetic Resonance Imaging) (7) DCT(Discrete Cosine Transform) (8) high pass filters (9) lossy compression,3. Translate the following sentences into English. (1) 人眼具有这样的特性: 图像出现在视网膜上会保留几毫秒, 然后消失。 如果一个图像序列以50幅图像每秒逐行扫描显示, 人们就不会觉得看到的是一幅离散的图像。 所有视频系统都是利用这一原理产生运动画面的。 (2) 所有压缩系统均需要两种算法: 一个在信源压缩数据, 一个在信宿对数据进行解压。 在文献当中, 这两个算法分别被称做编码算法和解码算法。,(3) JPEG(联合图像专家组)用于压缩连续色调的静止图像(例如照片)。 它是ITU、 ISO和IEC共同支持的图像专家开发出来的。 (4) 宽高比是图像的宽度和高度之比。 常规电视的宽高比是43。 高清晰度电视使用169的宽高比。 摄像机常用1.851或者2.351的宽高比。 (5) MPEG代表运动图像专家组, 它是对数字压缩格式视听信息(如电影、 录像、 音乐)编码的一组标准的统称。 与其他视频、 音频的编码格式相比, MPEG的主要优势是相同质量下的文件要小得多。 这是因为MPEG使用了非常复杂的压缩技术。 ,4. Answer the questions. (1) What is the function of subtraction of images? (2) What is the function of warping? (3) What is dilation? (4) Compare lossless compression to loss compression.,Passage B Compression/Decompression Techniques Numerous methods have been developed for the compression of digital image data. One of the principal drivers for this development is the television industry where quality image data must be transferred to receivers using relatively simple equipment. The development of high definition television is further focusing the attention of industry and university scientists toward problems of data reduction and digital transmission. The principal evaluation criteria for the analysis of compressed versus uncompressed imagery is whether a person can tell the difference between the images. A more implemental measure is the Root Mean Square (RMS) error between the original image and the image that has been compressed. Compression rates may be generated by determining the size of the compressed image in terms of number of bits per image pixel for the original image.1,Here we only considers compression of single high resolution multi-spectral images. Higher compression rates will be achieved in a motion sequence where frame to frame variations may be quantified and only the changes from a reference image need be coded.,There are two general types of compression: (1) loss-less, and (2)loss. Loss-less compression means that you can achieve a certain compression factor and be able to exactly reproduce the original image. Loss compression on the other hand allows some loss, but has the potential for much higher compression rates. No matter what technique that you use, the exact rate is very dependent on the complexity of the image that you are analyzing. For example, the normal best that can be achieved with loss-less encoding in a rate of 2 bits per pixel. In fact, for some Land-sat scenes with urban areas and many small farms, the factor of 2 bits per pixel may not be able to achieved. The same technique applied to a Land-sat image of the mid-west where large fields occur and few shadows exist in images might produce a much better compression.,One loss-less technique is known as run length encoding. The compression algorithm processes each line of input imagery looking for regions in which data values are the same. If ten pixels in the original image have a value of 10, then the same data may be represented as a data value, 10, and a multiplier saying how many times the value is repeated before a changed value. Huffman encoding follows a similar process. These loss-less techniques are generally called entropy coding techniques, and have application in document imaging, desktop publishing, and GIS. It should be noted that entropy coding does not work exceptionally well in the representation of remote sensing images.,In remote sensing imagery it is well known that there may be significant correlation between different bands of multi-spectral data. In image processing, a procedure called principal components has been designed to identify correlation between image bands and to create a new set of transformed bands that represent a new color space in which the new image bands are uncorrelated.2 The procedure also provides a measure of the percent of the original variation present in the original image as found in each of the new transformed bands. For Land-sat TM data, three to four of the transformed images represent 98 percent of the variance in the original images; therefore, a compression factor of 2 would be achieved with little loss.,Another type of transform coding does not involve a rotation of the color space, but instead represents images in terms of spatial frequency of certain base functions. Fourier transforms map an image into a spatial frequency image base on sin and cosine functions. A fast computer implementation of the Discrete Fourier Transform (DFT) is known as a Fast Fourier Transform (FFT). Discrete Cosine Transforms (DCTs) map the same image to a spatial frequency image based only on the cosine function. Each pixel may be represented by a series of trigonometric functions and coefficients derived from the images. If all terms of the transforms trigonometric functions are used, compression is minimal. As more terms are deleted, compression goes up, but the resulting compressed image develops certain artifacts of the procedure.,Vector Quantification (VQ) is a type of encoding that defines a vector representation of non-overlapping area blocks within an image. A vector consists of values representing the data values for each pixel within the region. Using these vectors, clusters of vectors are derived using a derived spectral distance measure.3 A codebook consisting of the clustered vectors is stored, representing the characteristics of the image. This process is numerically intensive and may be iterative. The decoder for VQ takes an image vector and compares it to stored vectors in the codebook. A selection is made based on minimization of a distortion function between the new vector and the codebook. The VQ technique can generally achieve compression ratios of between 20 to one and 35 to one with little observable distortion.,The VQ technique is an example of asymmetric compression in that considerably more time is spent deriving the codebook than in decompressing via a codebook lookup. Since different images may have different characteristics, a robust codebook is necessary to successfully code and decode Land-sat and other satellite images. A VQ technique using between channel correlation as well as spatial correlation achieves higher compression rates with lossless than independent band VQ.,The last type of compression considered is fractal compression. Based on Mandelbrot sets which take advantage of a self similar, scaling independent, statistical feature of nature, fractal compression and decompression involves a clustering approach to finding regions which exhibit the same characteristics as a sample region without regard to rotation and scale. Regions within an image are related to numerous other regions within the same image, with this duplication of information being the basis of the compression potential. Fractal compression can achieve compression ratios of up to 80 to one with only moderate loss of information. The fractal technique, like the VQ technique is also asymmetric. Hardware implementation of the decompression of fractal images has achieved real-time rates.,NOTES 1 Compression rates may be generated by determining the size of the compressed image in terms of number of bits per image pixel for the original image.压缩比可以通过原始图像每像素所需压缩的比特数来确定。 in terms of意为“根据”。,2 In image processing, a procedure called principal components has been designed to identify correlation between image bands and to create a new set of transformed bands that represent a new color space in which the new image bands are uncorrelated.在图像处理过程中，一个称为主分量的过程被设计用以确认图像频谱之间的相关性，并产生一组新的变换频谱，其中新图像频谱互不相关地表示一个新的彩色空间。 that引导定语从句修饰bands。in which引导定语从句修饰space。,3 Using these vectors, clusters of vectors are derived using a derived spectral distance measure. 使用这些矢量，用一个导出的频谱距离可推导出矢量簇。 第二个using相当于by using。,EXERCISES 1. True/False. (1) Loss-less and loss are two general types of compression.( ) (2) No matter what technique that you use, the exact compression rate is not very dependent on the complexity of the image that you are analyzing.( ) (3) The decoder for VQ takes an image vector and compares it to stored vectors in the codebook.( ),2. Fill in the blanks. (1) Loss-less compression means that you can achieve a certain and be able to exactly reproduce the original image. (2) A fast computer implementation of the Discrete Fourier Transform is known as a . (3) Vector Quantification (VQ) is a type of encoding that defines a of non- overlapping area blocks within an image. (4) The fractal technique, like the VQ technique is also . (5) compression can achieve compression radios of up to 80 to one with only moderate loss of information.,3. Chose the best answer. (1) What kind of compression algorithm processes each line input imagery looking for regions in which data values are the same? a. run length encoding b. Vector Quantification c. fractal compression (2) Which type of compression take advantage of a self similar, scaling independent and statistical feature of nature? a. run length encoding b. Vector Quantification c. fractal compression,Passage C Television A television picture is built up gradually by moving a spot of light across and down a screen in a raster pattern. The video signal causes the brightness of the spot to vary in proportion to the intensity of light in the original image. The movement of the spot across the screen is controlled by the line scan signal. Each time the spot reaches the right side of the screen, it is blanked and moved rapidly back to the left side ready to start the next line. This rapid movement back to a starting position is known as fly-back. Each complete image or frame requires a minimum of 500 lines to give a picture of acceptable quality. The present European TV system uses 625 lines per frame.,The movement of the spot down the screen is controlled by the field scan signal. When the spot reaches the bottom of the screen, it is blanked and moved rapidly back to the top of the screen. The frame must be scanned at least forty times per second to prevent the screen from flickering. The present European TV system has a frame scan rate of 50Hz. The video signal contains line and field sync pulses to make sure that the TV receiver starts a new line and a new frame at the same time as the TV camera.,To allow the video signal to be transmitted using a smaller range of frequencies, each frame is transmitted in two separate halves, known as fields. The first time the spot travels down the screen it displays the first field, which consists of the odd-numbered frame lines. The second time the spot travels down the screen it displays the second field, which consists of the even-numbered frame lines. Combining two fields in this way is known as interlacing. Although the fields are displayed one after the other, it happens so quickly that the human eye sees them as one complete picture.,In Europe, the USA, and Japan, the race is on to produce a new generation of television sets. These new sets will be larger than todays models, possibly with 100-centimetre flat screens. Picture quality will be excellent, crisp, and without flicker, as good as those we are used to seeing in the cinema. Sound quality too will be superb, thanks to digital multi-track transmissions. By the turn of the century such sets may be offering programs in a choice of languages as they will be equipped with eight sound tracks.,In Europe, the term HDTV is used. In the USA, the more generic term ATV, Advanced Television, has been adopted. The Japanese, who were the first to start work on the new technology, in 1974,called their system Hi-Vision. Whatever name is used, these new sets share certain features. The picture is displayed using more lines per frame. This means that they provide clearer, more detailed, high quality images. The picture can be displayed on large, wide screens which are flicker-free. They also provide very high quality three-dimensional sound output.,A wider range of frequencies can be used to transmit each HDTV channel. This is because they can be transmitted at high frequencies which are virtually unused at present. These wide frequency ranges make it possible to transmit digital, rather than analogue signals. Digital processing can then be used in the receivers to provide almost per
展开阅读全文