Question

Related to Image Classification

Answer 1

Answers:

(1) Train the classifier.

(2) True

(3) Image Pre-processing

(4) Weakly Supervised Learning Algorithm

(5) SIFT (or SURF)

(6) True

(7) True

(8) True

(9) True

(10) Decision Tree Classifier

(11) Softmax

Explanations:

(1) In supervised learning, we have given labels (y) and we have input examples (X) which we need to classify. In Keras or in Scikit-learn, we have a function fit(X, y), which is used to train the classifier. In other words, you have to train the classifier by using the incoming inputs (X) and the labelled outputs (y). Hence, the correct answer is: The fit(X,y) is used to train the classifier.

(2) This statement is primarily talking about the PCA, which stands for "Principal Component Analysis." It is a technique or method used to compress the given data, which is huge, into compact representation, which represents the original data. That representation is the collection of PCs, which are Principal components. PC1 represents the axis that covers the most variation in the data. PC2 represents the axis that covers the variation less than that of in PC1. Likewise, PC3 represents the axis that covers the variation less than that of in PC2, and so on. Therefore, it's true that the variation present in the PCs decrease as we move from the 1st PC to the last one.

(3) Image pre-processing is the phenomenon (or you can say the set of operation) used to improve and enhance the image by targeting the distortions within the image. That distortions are calculated using the neighbouring pixels of the given pixel in an image. Hence, the correct answer is Image pre-processing.

(4) SVM stands for State Vector Machine. It is basically a classifier, which is used to classify different (given) classes with precision. In simple terms, you can say that it is an algorithm, which is partially based on the given labeled data to predict the inputs. In technical terms, we call it weakly supervised learning algorithm. Hence, the correct answer is: Weakly supervised learning algorithm.

(5) There are many algorithms out there to detect the matching regions within two images. SURF (Scale Invariant Feature Transform) and SIFT (Speeded up Robust Feature) are two algorithms that can be used for matching patterns in the given images. Hence, you can choose any one of the two: SURF and SIFT.

(6) Indeed. Higher the accuracy is, better the classifier will be. However, there is a problem of overfitting that occurs when the accuracy of the classifier is way too high. Nevertheless, mostly, the classifier is better when there is higher accuracy. Hence, the correct answer to your question is true.

(7) True. Gradient descent is the process used to tune the parameters of the given neural network in order to decrease the error and increase the accuracy of the classifier. It calculates and fine-tune the parameters from the output to input direction by taking the gradient of the error function (sometimes called the loss function), which is the technique called backpropagation. Hence the correct answer is true.

(8) True. As explained in the part (5), SIFT which is called  scale-invariant feature transform, is an algorithm used to detech the features or the matching regions within given images. Hence, it's true that scale-invariant feature transform can be used to detect and describe local features in images.

(9) True. Clustering is indeed a supervised classification. In clusterning, we use graphs, which contains different data points in the form of clusters, to visualize the data as well. Imagine we have 7 fruits, out of which we know 6 of them, and we have to predict the 7th one. Let's say, 3 are apples and 3 are oranges. The set of apples is one cluster, and the set of oranges is another cluster. Now if we predict the 7th one by using the clustering technique, under the hood, that technique/algorithm will first train the model using the 6 fruits, which are known and then predict the 7th fruit. This kind of technique is a supervised learning, and hence, we can say that clustering is a supervised classification.

(10) In machine learning, Decision Tree Classifier is used to predict the value of the given input based on various known input variables. In this classifier, we can use both numeric and categorical values to get the results. Hence, the correct answer is Decision Tree Classifier.

(11) Softmax is the function which is used to convert the K-dimensional vector into the same shaped vector. The values of the Softmax function lies between 0 and 1, and it is primarily used as an activation function in a classification problems in neural networks (or deep neural networks). Hence, the correct answer is Softmax.