FYP progress 3

Haiz ...after i study my fyp for so long .. today only realise there is some misunderstanding of my concept for my fyp ....thx to my supervisor's class mate last time, which is also a lecturer in UNITEN. She told me the real concept and with it, i think i am proceding to the right direction...

The missunderstanding of the concept is in reducing the coefficient. Today the lecturer told me, the coefficient is reduce by multiple with the identity matrix (1 row and M coloum),

T = dctmtx(8) %generate transform matrix
B = blkproc(I,[8 8],'P1*x*P2',T,T') %Apply T across the image in 8*8 blocks
B2=B*eye(255, 1) % reduce the coefficient

S0 B2 will be some value in matrix (255*1) .... so this will be the input which will be choosen in future to be trainned....

*****think at last there is some progress in my FYP ... . if not sure stuck in progress report 2

忙里偷闲

最近可以算是近来比较空闲的几天，就分享几个烂笑话，让大家忙里偷闲
1）
如何分辨章鱼的手和脚？我们常说章鱼有八只脚，可是也有人说是八只触手。
那到底章鱼有几只手，几只脚？该如何分辨手和脚呢？大家想想后，再看答案吧
方法一：先抓一只章鱼，再用铁锤往牠的头打下去，章鱼会去摸自己头的就是手，其它的就是
脚啰。
方法二：如果是母章鱼的话，先假装要去强暴牠，它猛说不要时一直挥动的那几只就是手，其
他几只夹得紧紧的就是脚。
方法三：放个屁给牠闻，会摀住鼻子的就是手，其它的就是脚。
方法四：给牠一台计算机，放在键盘上的就是手，盘在椅子上缩起来的就是脚。
方法五：夏天到了，会抓香港脚的就是手，被抓的当然就是脚啰！

2）有一天，阿荣去北京办事，晚上到一家旅馆投宿。
阿荣就叫旅馆妈妈生来说：妈妈生借问一下，妳这有没有小姐阿？
妈妈生就连忙说有阿有阿，于是她就把旗下所有小姐叫进来让阿荣挑，阿荣看了看就挑其中一
个陪宿。
晚上就跟她xxxx后问他要多少钱一晚，小姐说1晚1千元阿，
阿荣就说：妳服务真周到，我给妳1万元好了。
小姐觉得真是遇到贵人了，就跟他说如果明天有需要可以打1380******给她。
隔天阿荣又叫那小姐来，她就比昨天更卖力的服务阿荣，阿荣又给她一万元。
小姐就问他要住几天阿，阿荣说住3天啦，第三天也麻烦妳服务一下。
第3天小姐就很开心更卖力的服务下去，阿荣依然给了他一万元。
那小姐收了前后不禁叹息说：很久没遇到像你那么好的人客了，对啦，先生是从哪里来的？
阿荣就说：俺从山东的啦...
小姐说这么巧唷，俺也是那里人唷，你该不会是上杨村的人吧？
阿荣就说：是阿！是阿！我就住村尾阿！
小姐说：真巧啊，我刚好住村头，你来北京作什么阿？
阿荣就说................
"没什么啦，我出差来北京，你妈托俺顺便捎3万元给妳”

3）有一天，一个乡下佬去看电影。
买了票之后，走进电影院，可是过了一会，又气呼呼走出来再买了一张票，走进电影院。
售票小姐觉得很奇怪,可是还是卖给他。
结果过了一分钟，又见那个乡下佬发疯似的冲向售票口，再买了一张票。
这次售票小姐就问他说「不是已经买了票吗，干吗还要再买啊？」
乡下佬很生气的说「我怎么知道，每次我一走进电影院，就有一个小姐把我的票撕掉

My progress on my FYP

My progress on my FYP ~~ for those interested > . < '' my project can be divide into 4 part basically

input -->image processing --> Artificial neural network --> output

Here is part of my hard work on the image processing part, it is done through Discrete Cosine transform. Here is the theory on how it working again . Firstly, the signal originally is in spatial domain. By DCT, we transform it to frequency domain. In frequency domain, the signal is in spectrum form where most of the information of the image is in low frequency. Our eyes also not sensitive to high frequency information, so theoretically we can filter the high frequency out without affect the image. THen IDCT is done to transform it back

so this is the result of filter out the high frequency, because the photo are not really clear after upload, so we cant really c the different at here . But if it is in the computer, the first 1 is the original one and the other is 50% filter on row and coloum ,75% filiter on row and coloum and 87.5 filter on the row and coloum information. The more we filter out, the image will become more "blur"

This is the same image in grayscale >.<

after DCT and IDCT, the feature needed to be trained can be reduced

** will it only make ppl feel ever blur ???

My Final Year Project tittle

Posture Recognition using RBF and MLP Neural Network

*now maybe change to defective tin a.k.a tin rosak recognition =.= '''

Some brieft introduction on field of my fyp [boring literature review]

Artificial neural networks can be most adequately characterized as “computational model” with the properties such as the ability to adapt and learn, to generalize, to cluster or organize data where all the operations are working based on parallel processing. For the above properties, there are some other existing non neural models can be attributed. Artificial neural networks are proven to be a better approach for certain applications.

The utility of artificial neural network models lie in the fact that they can be used to infer a function from observations. It comes in hand in applications where the complexity of the data or task makes the design of such a function by hand impractical. Artificial neural network are applied in broad categories from functional approximation (time series prediction, fitness approximation and modeling), classification (pattern and sequence recognition, sequential decision making), data processing (filtering, clustering, compression) and robotics (directing manipulators) ^[1]

RBF network (Radial basis function) and MLP network (multi-layer perceptron) are commonly involved in artificial neural network classification application. A RBF network is an artificial neural network that uses radial basis functions as activation functions. Generally, it consists of two layers of processing. The input is mapped onto each RBF in the hidden layer. The RBF is usually a Gaussian. A MLP is a feed forward artificial neural network model that maps sets of input data onto a set of appropriate output. It uses there or more layers of neurons with nonlinear activation function.

MLP


RBF

I think most of the people wont know what it is and wont interested to know it also ..... maybe a real life example will be easier

When we taking photo with camera, there is a function as smile detection or face detection. It will capture the motion of the face and snap when there is a smile or focus on the face (theoretically >.<). Basically, the input is the image capture by the camera and the input have been trained with ANN(RBF,MLP, SVM etc) to perform the function. The neuron is train with the image such as face or non face before its perform the function . The theory of how my FYP working is similar to this one....

* i doubt do people can understand this o not ???