Image processing Part 3 :

Image preprocessing

According to Andrew Ng data is the new electricity. And it is important to have regular electricity for us to make our life easier. Similarly, the data needs to be qualitative and clean. Building model can be done using others model with same parameters, layers, etc. but actually what matters most is its data quality.

Image processing is the series of process that involves transformation, rotation, resizing, color transformation. etc.

WHY

Real world data are messy and wrong have different size color. A model can understand only if we train with common characteristics. In case of different features, there is no chance we can build model.

what if this happens ?

Generally when we try to predict for motorbike, it will prompt us………

Techniques to standardize images

a) convert color images to grayscale to reduce computation complexity

<color doesnot provide much information rather increases space complexity and time complexity too>

• **** case ::::: color important:::::::::********

eg. building a model to identify skin rashes,

eg. lane detection

\

b) Standardize Images

resize the higher dimensional images.

c) Data Augmentation

using math for transformation, rotation, de-colorization

Data augmentation

d) Other

Sometimes background may be required to be removed. Some require brightening or darkening of the image depending upon the problem.

No Free lunch Theorem

David wolpert’s paper ‘No free lunch theorem for optimization’

There is no any recipe that fits all the models.

We must assume about the dataset and the problem we’re trying to solve. Classicial model requires high level of data preprocessing task . But new deep learning approach has helped minimize the data pre-processing tasks.